PRIVATE SCHOOLS IN THE SOUTH:  IS IT ABOUT EDUCATION? 
 
 
 
Except where a reference is made to the work of others, the work described 
 in this thesis is my own or was done in collaboration with my advisory 
committee.  This thesis does not include proprietary or classified 
 information. 
 
 
 
______________________________________________________ 
Rachel Wyatt 
 
 
 
Certificate of Approval: 
 
 
_________________________                       __________________________ 
Curtis Jolly                                 Greg Traxler, Chair 
Professor                                  Professor 
Agricultural Economics and                              Agricultural Economic and 
Rural Sociology                             Rural Sociology 
 
 
_________________________                  __________________________ 
Henry Thompson                           Henry Kinnucan 
Professor                                Professor 
Agricultural Economics and                    Agricultural Economics and 
Rural Sociology                            Rural Sociology 
 
 
 
________________________ 
Stephen L. McFarland 
Acting Dean 
Graduate School 
  
 
PRIVATE SCHOOLS IN THE SOUTH: IS IT ABOUT EDUCATION? 
 
 
Rachel Wyatt 
 
 
 
A Thesis 
 
 
Submitted to 
 
 
the Graduate Faculty of 
 
 
Auburn University 
 
 
in Partial Fulfillment of the  
 
 
Requirements for the 
 
 
Degree of 
 
 
Master?s of Science 
 
 
 
Auburn, Alabama 
May 11, 2006
PRIVATE SCHOOLS IN THE SOUTH:  IS IT ABOUT EDUCATION? 
 
 
 
 
Rachel Wyatt 
 
 
Permission is granted to Auburn University to make copies of this thesis at its 
discretion, upon request of individuals or institutions and at their expense.  
The author reserves all publication rights. 
 
 
 
 
_____________________________ 
Signature of Author 
 
 
 
_____________________________ 
Date of Graduation 
 
 
 
 
iii 
 
VITA 
 
 Rachel Karen Wyatt, daughter of Terry and Karen Wyatt, was born April 
12, 1982, in Birmingham, Alabama.  She was reared in Childersburg, Alabama, 
on a family farm that grew cotton and corn.  She graduated from Childersburg 
High School in Childersburg, Alabama, in 2000.  In August of 2001, she entered 
Auburn University and received a degree of Bachelor of Science in Agricultural 
Economics.  She entered the Graduate School of Auburn University, in August of 
2004.  While fulfilling the requirements for a Master of Science degree in 
Agricultural Economics, she was employed as a Graduate Research Assistant in 
the Department of Agricultural Economics and Rural Sociology under the 
direction of Dr. Greg Traxler. 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
iv 
 
v 
THESIS ABSTRACT 
PRIVATE SCHOOLS IN THE SOUTH:  IS IT ABOUT THE EDUCATION? 
 
Rachel Wyatt 
Master?s of Science, May 11, 2006 
(B.S., Auburn University, 2004) 
 
89 Typed Pages 
Directed by Greg Traxler 
 
The question of whether or not private schools provide a better education 
than public schools has been the study of many studies. The question is 
important because of the large amount of tax money that is devoted to public 
schools, and because of the impact that elementary and second education has on 
the workforce and the economy. The findings in past studies have been mixed. 
Some studies, such as Figlio and Stone (1999), have found that private schools 
provide a better education. They argue that private schools respond to 
competition in a way that public schools do not, and their analysis show that 
private school students perform at higher levels at lower expenditures per pupil, 
 
vi 
especially in religious schools. Other studies have failed to find any effect of 
either private versus public school attendance or of difference in funding levels 
(Hanushek, 2001). A number of studies have examined the decision to attend a 
private school. Figlio and Stone (1999) found that the larger the fractions of 
minorities in the county, the more likely white students were to attend a private 
school. Long and Toma (1988) test many economic characteristics of the 
household that influence private-public school choice.  The incorporate 
characteristics such as, household income, ethnicity, education and age of the 
head of the household head.  In the South, issues of segregation and school 
quality are closely related to the formation of private schools. The two decades 
following Brown vs. Board of Education witnessed the founding of private 
academies in the majority of counties in Alabama, Mississippi and Georgia. 
Although officially open to all ethnic groups, these academies have resulted in 
the de-facto segregation of many rural areas. In some areas, the student bodies of 
the academies remain almost totally white, while public schools are dominantly 
African-American. One important consequence of this dual school system is that 
local school funding referendums often fail because those residents whose 
children attend private schools are unwilling to increase their own taxes to 
improve public schools. In smaller rural communities, this often results in two 
inadequately funded schools ? one private academy with one or two hundred 
students dependent on tuition funding, and a public school funded with 
vii 
inadequate local tax rates.  In testing for the effect of private schooling, previous 
studies have not distinguished among the various classes of schools. We 
hypothesize that there are significant differences between older, more 
established private schools (for example a 100 year old Jesuit school), and a small 
southern ?academy? to subvert integration. We use dummy variables to capture 
this difference, allowing the effect of attendance at a small academy to differ 
from the effect of a public school. The null hypothesis to be tested is that 
established private schools and the ?segregation academies? have the same effect 
on Test Improvement and Income. If we fail to reject we will conclude that 
parents had the same motivation to send their children to attend southern 
private academies as parents who sent their children to other schools had ? a 
better education. Rejection of the null hypothesis will lead us to conclude that 
there were different motivations. 
 
ACKNOWLEDGEMENTS 
 
 Special appreciation is extended to the author?s chairman, Dr. Greg 
Traxler, for his patience, counsel, and guidance throughout graduate study.  
Appreciation is also due to Dr. Henry Kinnucan, Dr. Henry Thompson, and Dr. 
Curtis Jolly for professional advice and assistance during the development of this 
thesis.   
A special acknowledgement is expressed to the Department of 
Agricultural Economics and Rural Sociology for the opportunity to pursue a 
graduate study and to receive a graduate research assistantship.   
 The author also wished to thank her parents (Terry and Karen Wyatt), 
sister (Lauren Naile), and Kevin Holland for their advice, comfort, support and 
encouragement given during graduate study.                                                    
 
viii 
 
 ix
Style Manual or journal used: Journal of American Agricultural Economics 
Association 
Computer software used: Microsoft Word, Microsoft Excel, and STATA 
statistical software 
  
 x
TABLE OF CONTENTS 
LIST OF TABLES............................................................................................................. xi 
INTRODUCTION .............................................................................................................. 1 
General Objective ....................................................................................................... 2 
LITERATURE REVIEW ................................................................................................... 3 
     Private School Education History.................................................................................. 3
     a. History of school integration.................................................................................. 3
     b. How Private Schools developed in the South........................................................ 6
     Public and Private School Education............................................................................. 7
     a. Benefits of private schooling ................................................................................. 7
     b. Benefits of public schooling ................................................................................ 10
     c. Decisions determining the choice between public and private school attendance13
     Per Capita Income vs. Education................................................................................. 14
     a. Determinants of per capita income ...................................................................... 14
     Public and Private School Education........................................................................... 20
    a. Public vs. Private Education ................................................................................. 20
     Economic Factors that Affect Education ..................................................................... 25
    a. Effects of dropout rates in High School................................................................ 25
    b. Effects of Educational Spending........................................................................... 28
    c. Differences in student performance between race ................................................ 34
     Economics Factors that Effect Per Capita Income ...................................................... 36
     a. Education and Per Capita Income........................................................................ 36
MODEL AND VARIABLE ANALYSIS......................................................................... 43 
     Student Performance Model Description..................................................................... 44
      Variables: ................................................................................................................ 44
      The School choice Model Description........................................................................ 46
      Variables: ................................................................................................................ 46
    Earnings Model Description ......................................................................................... 47
          Variables: ................................................................................................................ 48
Dataset............................................................................................................................... 49 
Summary Statistics............................................................................................................ 51 
MODEL RESULTS.......................................................................................................... 54 
Variables ....................................................................................................................... 54 
    Tests for Models ........................................................................................................... 55
Student Performance Model Results............................................................................. 64 
Income Model Estimation............................................................................................. 70 
SUMMARY AND CONCLUSION ................................................................................. 73 
BIBLIOGRAPHY............................................................................................................. 75 
 xi
LIST OF TABLES 
Table 1 Definition of all variables????????????????????58-59 
Table 2 Summary Statistics of all variables used in the models??????????52 
Table 3 School Choice Model estimation results??????????????.60-62  
Table 4 Student Performance Model estimation results???????????...66-67 
Table 4a Student Performance Model estimation results  
(Including variable PRIV)??????????????????????....68-69 
Table 5 Income Model Estimation results?????????????????71-72 
Table 5a Income Model Estimation results (including variables UNDGPA)?????72 
 
 1
CHAPTER I 
INTRODUCTION 
 
This paper discusses the role that private schools play when parents choose a 
school for their children, with special reference to the South.  It presents evidence to 
support two main questions.  First, are parents sending their children to a private school 
based on factors other than school quality?  Second, does attending a private school 
anywhere in the United States affect annual income?  Private schools in the South have 
grown in importance since 1960.  Because of the typically large areas covered by the 
school districts in the South, private schools have offered white families a means of 
avoiding integration particularly in counties with high minority concentrations.  In those 
counties the rate at which whites enroll in private schools tends to rise with the 
percentage of students who are non-white (Ashley 2001, p. 70).  This study examines the 
role private schools play with respect to school segregation.  Therefore, do parents in the 
South choose to send their children to a private school based on only academic 
performance?  As annual income increases private schools become financially feasible 
for more families.  This is only one of the reasons why it is important to learn about 
patterns of private school enrollment. 
The purpose of this study is to determine the factors that parents consider when 
choosing a school.  Educational outcome in this study is examined through an analysis of 
standardized achievement test scores for 1988 -1992, using the average of the 
standardized test scores in the subject areas of reading, math, science, and history.  
 2
Annual income in this model is also represented by the students income at age 25 in 
1999.  The purpose of the income analysis is to demonstrate how education can affect 
income.  Does attending a private school help increase annual income and provide 
resources that public schools cannot?  This paper provides the supporting evidence to 
respond to these concepts. 
General Objective 
To determine whether parents choose a school for their children based on factors 
other than academic performance compared to the alternative that parents choose a school 
based on only academic performance. 
 3
CHAPTER II 
LITERATURE REVIEW 
 
There is a growing literature on the current trends in income and education.  
Much of the existing literature on the effect of educational attainment on income focuses 
on high school non-completion rates, expenditures, student teacher ratios, and the 
benefits of public versus private schools.  School studies have compared income to high 
school drop out rates.  Other studies have discussed the determinants of per capita income 
and how educational spending might affect it.  
Private School Education History 
a. History of school integration 
 One of the most significant events in Alabama education was the reversal of the 
?separate but equal? policy for schools for African American children in the state.  ?After 
a lengthy series of court battles to insure full status for black students in public 
institutions of higher education, the crucial case of Brown vs. Board of Education came 
before the United States Supreme Court? (Ashley 2001, p. 70).  On May 17, 1954 the 
court declared that segregation must vanish. The ?separate but equal? policy established 
by Plessy vs. Ferguson (1896) had officially come to a halt.  Recognizing the 
revolutionary nature of its decision, the Court in a case the following year indicated that a 
gradual approach to desegregation would be acceptable.  The majority of the Border 
States took relatively early action to carry out the Supreme Court?s decision, but in 
 4
Alabama where African Americans were numerous, the decision met with much 
resistance.  By the end of 1957 nine of the 17 Border States and the District of Columbia 
had begun integration of their school systems.  Another five states had some integrated 
schools by 1961.  The states mostly fell back on substitute measures or on pupil-
placement laws, which assigned students to schools ostensibly on nonracial grounds 
(Clarke 1989, p. 270).  Forced integration led to much violence.  Groups of whites 
gathered in large numbers in some places in Alabama to prevent the integration of 
schools (Clarke 1989, p. 270).  There were many events where the white people would 
torment the African American students.  For example, when the state was ordered to 
admit Autherine Lucy, one of the first African American applicants, to the University of 
Alabama, some irresponsible students and townspeople of Tuscaloosa resorted to 
violence to prevent her from remaining at the University (Clark 1989, p. 270). 
The dual system in Alabama was further dismantled by the Title VI of the 1964 
Civil Rights Act, which banned discrimination in all federally aided projects and 
programs throughout the country (Clarke 1889, p. 271).  This act was effective since 
virtually every school district in the U.S. received some type of federal aid.  Alabama?s 
two major universities, most of the smaller state colleges, and the larger public school 
systems were integrated by 1965 (Clarke 1889, p. 272).  There were various Supreme 
Court decisions in the early 1970?s, such as ?Singleton vs. Jackson Municipal Separate 
School District which set forth black and white faculty requirements in desegregated 
schools and Swann vs. Charlotte-Mecklenburg which declared that student transportation 
must be used whenever necessary to eliminate racial isolation of students.? (Clarke 1889, 
p. 273).  However, desegregation of the schools systems in Alabama was not easy.  Much 
 5
                                                
of the focus of desegregation efforts in the state was in small rural districts where 
resistance to desegregation was often the most determined and violent.  In these rural 
districts, which is relatively easy to desegregate in the structural sense, were often the 
first to abandon the formal trappings of racially separate schools.  Attention shifted to the 
urban school systems not only of Alabama and the South, but of the whole nation.  In 
these systems the complexities of size, geography, transportation, and hypocrisy made 
school desegregation an issue, and demonstrated that resistance and violence were not 
peculiar to any one area of the United States.   
White flight
1
 has also affected education.  American cities affected by white flight 
also witnessed growing disparities in the quality of education.  Busing and desegregation 
orders in education have in some cases led to further, non-geographical white flight:  out 
the public school systems, which are subject to desegregation order, and into private 
schools, which are not.  For instance, in 1970, when a federal court ordered desegregation 
of the public schools in Pasadena, California, the proportion of white students in those 
schools reflected the proportion of whites in the community, 54 percent and 53 percent, 
respectively (Ashley 2001, p. 70).  After desegregation began, a large number of whites 
in the upper and middle classes who could afford private schooling pulled their children 
from mixed public schools.  As a result by 2004 Pasadena was home to three private 
schools, which educated one-third of all school-aged children in the city, and the 
proportion of white students in the public schools had fallen to 16 percent (Ashley 2001, 
p. 72).   
 
1
 White flight is a colloquial term for the demographic trend of white people, generally but not always 
upper and middle class, moving from increasingly and predominately non-white areas.   
 6
b. How Private Schools developed in the South   
The enforcement of school desegregation often is called ?school consolidation or 
reorganization.?  Brown I and II, initially sought to merge or consolidate the ?state 
sponsored? dual school systems of the United States. The explicit mandate of Alexander 
v. Holmes (1969) to desegregate ?at once? applied to every school district in the country.  
The 1969 ruling bucked the ?with all deliberate speed? language of the past 15 years and 
said that every school district in the South would remain under court order until ?unitary 
status? was achieved as measured by statistical means.  Through the leadership of 
Governor Wallace and with no shortage of public outrage in the Deep South, the shift 
from in practice dual systems resulted in a new dual school system characterized by the 
sudden emergence of exclusively white segregation academies (Ashley 2001, p. 65).  The 
all-white schools placed less emphasis on quality education and more on maintaining 
racial homogeneity in all things education.  For the majority of Alabama, the public 
schools never fully desegregated due to massive white-flight to private schools.  As 
Clotfelter (2002) pointed out in his study of ?Private Schools, Segregation, and Southern 
States,? since 1960 ?the rate at which whites enrolled in private schools tended to rise 
with the percentage of all students who were nonwhite, increasing sharply in counties 
over 60% nonwhite.?   
The first segregation academy appeared in Prince Edward County, Virginia in 
1959.  The school is still in operation and has actually been integrated by minority-group 
students (Ashley 2001, p. 65).  The hard- fought school desegregation case in the county 
would serve as a last-resort model for southern communities and segregation academies 
that were willing to resist school desegregation by any means necessary.  However, 
 7
chances are if a private school was around in Alabama before 1954 it had little to do with 
race? but if the school has only been around since 1954 it has everything to do with 
race.  
John Herron?s (1977) study of the growth of private schools in Alabama showed 
that from 1967 to 1971 approximately 50,000 white students left the public schools 
(Ashley 2001, p. 66).  He found that there were 34 private schools operating in 1965 and 
109 operating in 1970- an increase of 220.6% in a five year period.  Walden and 
Cleveland?s study of changes in Alabama?s public schools also took notice of the 
educational ?white flight,? revealing that as the percentage of blacks in attendance at 
formerly all-white schools in district increased, the percentage of white students leaving 
the public schools increased correspondingly.  However, when black students reached a 
percentage of the total enrollment over 50%- 1 out of 5 white students left (Ashley 2001, 
p. 67).  By 1972, there were reportedly 140 segregated elementary and secondary private 
schools in Alabama with most enrollments below 200 students (Ashley 2001, p. 68).  
Public and Private School Education 
a. Benefits of private schooling 
 Neal?s (1998) review discusses the benefits of private schooling.  Neal discusses 
arguments about public and private school education through a literature review.  This 
paper attempts to summarize Coleman?s (1981) research and also to assess what has been 
learned since 1981.  Although many questions remained unanswered, one result seemed 
to be clear.  Black and Hispanic students in large cities often have the most to gain from 
private schooling.  Neal pays close attention to the literature on Catholic school effects 
because Catholic schools constitute a large and relatively homogenous group of private 
 8
schools (Neal 1998, p.80).  Neal examined how Coleman (1981) took on the task of 
evaluating the first wave of data, which collected achievement test scores for 
approximately 50,000 high school students who were in either their sophomore or senior 
year of high school.  This author argued that Catholic and other private schools are more 
effective institutions of learning than public schools.  Neal reports that Coleman?s study 
concerned the relative educational performance of public and private schools.  For 
example, Coleman established that in a population of students from similar backgrounds, 
private school students exhibit higher achievement and attainment.  He also found that 
African American students in larger cities perform better academically in private schools. 
His results illustrate that urban minorities attending Catholic secondary schools 
experience a 26-percentage point increase in the possibility of graduating from high 
school. Coleman also includes race and community type in his research.  Neal uses 
Coleman?s research to expand on his own research, by estimating the effect of attending a 
Catholic secondary school on the probability of graduation.  Neal believed that Catholic 
school (private school) attendance is beneficial to children and improves their academic 
future.   
 Figlio and Stone (1999) also determine that private schools provide a better 
education.  They believe that private schools respond to competition in a way that public 
schools do not.  They illustrate that private schools are superior to public schools in 
providing educational services.  FS state that private school students perform at higher 
levels at lower expenditures per pupil, especially students in religious schools.  They did 
find with statistical significance that private schools increase the probability that students 
will attend at least a two year college or some other academic institution.  Figlio and 
 9
Stone?s empirical results indicate that the higher the crime rate in the county, the more 
likely parents are to send their children to private schools.  They also found that the more 
concentrated the public schools in the county, the more likely parents are to send their 
children to private schools, mainly nonreligious private schools.  One important finding 
was that the larger the fraction of minorities in the county, the more likely that white 
students would be to attend a private school.  Figlio and Stone?s main empirical finding is 
that parents are responsive to the interactions between community characteristics and 
their own characteristics when considering a specific school for their children.   
Long and Toma (1988) discuss the main determinants of private school 
attendance.  They establish that public school enrollment is decreasing, so therefore 
private school enrollment is increasing.  They test many economic characteristics of the 
household that influence private-public school choice.  LT incorporate several 
socioeconomic characteristics of the household that influence private-public school 
choice.  Annual household income is expected to affect private school attendance 
positively.  Annual income in any year is an inexact measure of ?permanent? income.  
The authors include education and age of household head, along with homeownership, to 
proxy for permanent income.  They hypothesize that parents with a higher education are 
more likely to have a higher demand for private schools.  Demographic influences on the 
demand for private schooling are captured by including dummy variables indicating 
whether the household head is white or female.  They also considered the influence of 
school choice by regional and urban locations as well as southern central city and non-
southern central city using dummy variables (Long and Toma 1988, p.352).  Income is 
highly significant in LT?s model, implying that the demand for private schooling rises as 
 10
household income increases.  LT also estimate how central city public schools have a 
higher crime rate, which allows private schools to appear to be a better alternative (Long 
and Toma 1988, p 353).  They determined that public-private school choice is influenced 
by many demand factors such as income level, educational attainment, and race.  Long 
and Toma found that the private school attendance gap between low and high-income 
households and between white and nonwhites were smaller in 1980 than 1970 (Long and 
Toma 1988, p 356). 
b. Benefits of public schooling 
 Hanushek?s (1986) study examines research on the economics of education and 
schooling.  Hanushek discusses public school education policies.  He states that total 
spending on elementary and secondary public schooling is currently equal to about 4 
percent of the gross national product (Hanushek 1986, p.1143).  One important factor in 
public schools is the rise in per pupil expenditures in elementary and secondary public 
schools.  He illustrates how the student teacher ratio was greater in public schools than in 
private schools during the 1960s and 1980s.  He also demonstrates that more than half of 
public school teachers in 1983 held at least a master?s degree (Hanushek 1986, p.1145).  
Hanushek felt it was necessary to consider standardized test scores in his research.  Test 
scores might be a better indicator of learning in the earlier grades, where the emphasis 
tends to be more on basic cognitive skills of reading and arithmetic than in the later 
grades.  Hanushek establishes that more education and more experience on the part of the 
teacher costs.  Smaller classes (more teachers per student) should not have any influence 
on the improvement of an individual student learning.  Hanushek demonstration 
explained that school inputs do not have an affect on student improvement.  
 11
 Flower (1988) argues that tuition tax credits would provide stronger competition 
within the private sector.  However, Flower establishes that tuition tax credits actually 
decrease the quality of public education.  Flower discusses how parents have a negatively 
sloped demand curve for educational quality.  If public school quality is below standard, 
parents will be more likely to send their children to a private school.  On the other hand, 
if public educational quality exceeds their standards, parents will be more likely choose a 
public school for their children?s education (Flower 1988, p.92).  He believes poor 
quality is due to institutional problems rather than technological problems.  Flower 
illustrates how every child in the public school system receives the same quality, but that 
the parents can alter that quality by choosing to place them into a private school (Flower 
1988, p.94).  Flower describes how public schools are required to accept any child within 
that school district who wants to attend.  This is an important budgetary cost of the 
school?s quality.  Flower determines that the quality of education in the public system is 
low because decision makers are not confronted with the private costs that correspond to 
the school?s marginal costs.  
Belfield (2002) argues that in the educational sector, ?consumers? are 
parents/children and ?suppliers? are schools/districts, so more competition should 
translate into higher quality schooling and enhanced educational outcomes (Belfield 
2002, p.279).  The research reviewed here examined competition among U.S. schools and 
school districts.  It spans 1972 to 2002 and includes only studies with an explicit measure 
of competition.  Belfield answers the question of what ?competition? in the education 
sector mean.  He states that an education market exists whenever parents have a ?choice 
set? of providers.  In the U.S., parents may choose between 1) public and private schools, 
 12
2) among public school districts, 3) among public schools within a given district.  When 
there are more choices, there is more competition, but the presence of many choices does 
not necessarily ensure greater competition.  The second question Belfield examines is 
?does more competition raise test scores??   In many cases, tests were higher where there 
was more competition (Belfield 2002, p.280).  The third question discussed was whether 
more competition could improve education in other ways.  For example, if competition 
motivated schools to offer a better education, students may respond by remaining 
enrolled or by applying to college in greater numbers.  The fourth question asks about the 
reliability and validity of the evidence on competition.  Belfield found that where school 
quality is low, more families choose private schools, so it appears that competition 
influences school quality (Belfield 2002, p.281).  Overall, the evidence discussed in this 
study suggests that increasing competition-either intra-district or from private schools-
may raise effectiveness and efficiency of public schools, and may address other 
educational objectives. 
Chambers (1976) discusses how teachers? salaries vary across school districts and 
teacher characteristics.  He reviews researchers? work and demonstrates that the major 
impact of bargaining for teachers has been on a regional rather than on a district level 
(Chambers 1976, p.325).  Bargaining does have a substantial and significant effect on 
teachers? yearly wages. His focus is on the impact for individual public schools.  He 
discusses two major concerns for school decision makers.  One concern is the allocation 
of educational resources among the various educational inputs.  The second concern is 
the allocation of community resources between education and all other goods and 
services (Chambers 1976, p. 327).   Chambers found that bargaining power could affect 
 13
the allocation of resources in many ways.  The greater the relative bargaining powers of 
the teachers? union, the greater the extent to which the constraints will confine district 
decision making (Chambers 1976, p. 330).  As regional strength of the teachers? union 
increases, teachers? salaries will tend to increase in both bargaining and non-bargaining 
districts (Chambers 1976, p. 328).  The observed salaries of school personnel employed 
in district i are written as follows:  Wi = Wi + B1 fi +B2, where B1 measures the salary 
differential between bargaining and non-bargaining districts (referred to as the district 
effects of bargaining); fi is a dichotomous dummy variable which equals one if formal 
negotiations occurred in district i and zero otherwise; and BB
2
 measures the extent to 
which the salaries in nonbargaining districts exceed those that would have prevailed in 
the absence of bargaining in the region (Chambers1976, p.  328). Collective bargaining 
has a positive and statistically significant impact on teachers? salaries.  Teachers? salaries 
have increased by 5.7% and 12.2% in unified and elementary public school districts, 
respectively (Chambers 1976, p. 331).  Chambers shows how this only occurs in public 
school districts and not in private schools.    
c. Decisions determining the choice between public and private school attendance 
 A parent?s goal is to find a school that will meet their child?s needs.  There are 
many differences between public and private schools.  For example, public schools 
cannot charge tuition.  Public schools are funded through federal, state, and local taxes, 
white private schools are funded through tuition, fundraising, donations and private 
grants.  According to the National Association of Independent Schools, the median 
tuition for private day schools in the United States is close to $12,000 for grades 1 to 3, 
$13,000 for grades 6 to 8 and $15,000 for grades 9 to 12 (Belfield 2002, p.282).  The 
 14
median tuition for boarding school is $12,000 for grades 1 to 3, $27,000 for grades 6 to 9, 
and $28,000 for grades 9 to 12.  There are many differences between public and private 
schools.  Admissions are a major difference between public and private schools.  Public 
schools admit all children.  By law, public schools must educate all children, including 
students with special requirements, whereas private schools are selective.  They are not 
obligated to accept every child, and in many private schools admission is very 
competitive.  The government has certain guidelines that public schools must follow.  
Public schools must follow all federal, state, and local laws in educating children.  Such 
laws usually include specifics about funding, program development and curriculum.  
Private schools are not subject to as many state and federal regulations as public schools.  
Since private schools are funded independently, they are not subject to the limitations of 
state education budget and have more freedom in designing curriculum and instruction.   
Per Capita Income vs. Education 
a. Determinants of per capita income 
 Fullerton (2001) discusses the impact of education on per capita income.  
Fullerton compared the size of Texas counties to the level of per capita income in each 
county.  He finds that low high school completion rates have a significant negative 
impact on per capita income in the state of Texas.  Fullerton compares county dropout 
rates to total enrollment in each school.  He finds that reducing the dropout rate to a level 
proportionate with the rest of the state would increase per capita income by more than 
$2,600 in 1990.  Fullerton also states that greater public infrastructure investment to 
improve transportation and communication links with the rest of the state would reduce 
school dropout rates and increase income.  
 15
Cox (1995) compares per capita personal income to the roles of the women and 
their education in Kentucky.  Kentucky?s per capita income was about 80% of the 
national average in 1995. While per capita income in other Southern states, such as 
Georgia, and Louisiana, has increased relative to the national average, Kentucky?s fell in 
the mid-1980s before returning to around 80% in 1995.  Kentucky?s average income 
failed to rise at a rate that would move it toward the national average.  Increasing labor 
participation by females who have a low education level has a strong influence on the 
failure of earnings to rise as they take low-skill, low-paying jobs.  Cox suggests that 
unless the educational attainment of Kentucky?s adult population is improved, per capita 
income may not rise in the future. This suggests that one means of improving Kentucky?s 
per capita income performance is to improve the educational attainment of its adult 
population. 
 Ferguson (1997) examines Alabama?s per capita income and educational 
attainment in 1997.  Ferguson shows that Alabama?s per capita income in 1997 had fallen 
compared to the national average.  Alabama continues to lag behind the national income 
average.  Ferguson states that Alabama?s income gap is due to the type of industries that 
dominate Alabama?s rural industrial base.  Rural Alabama is dominated by low skill, low 
wage industries-characteristics of a poorly educated workforce (Ferguson 1997, p.1).  
Alabama is in fact located in one of the fastest growing regions of the country.  This 
suggests that the state should not continue to under-perform the nation as a whole.  
Ferguson?s study demonstrates that educational attainment in the state has increased in 
the last 30 years, but has not caught up to the national average.  In 1998, 78.8% of adult 
Alabamians had at least a high school education compared to 82.8% of the rest of the 
 16
United States (Ferguson 1997, p.2).  Ferguson compares educational attainment in both 
rural and urban counties.  He found that the average educational attainment of adults in 
the urban counties is significantly higher than that of the adults in rural counties.  
Therefore, Ferguson implies that rural counties in Alabama ignore the quality of 
education.  He also questions why, in times of need, public tax referendums are 
frequently defeated in urban schools.  
Rupert and Schweitzer (1996) discuss how the decision to further one?s education 
is based on expected future financial gain. An individual must compare the expected 
return less the cost of that education to the net return to no further education.  Economists 
measure the return to education using an empirical earnings function.  Some 
specifications discussed in this article were used to measure wage differences among 
occupations, races, and sexes.  They examine a standard empirical earnings function, 
concentrating on the increase in income due to more education (Rupert and Schweitzer 
1996, p. 1101).  Several empirical and conceptual issues were addressed in this article.  
One issue was how education should enter into an analytical framework, so that the 
return to schooling can be estimated properly (Rupert and Schweitzer 1996, p. 1102).  
The second issue was separating the return to education from other factors, such as 
experience in the workforce (1102).  It was found that more people between the years of 
1963 and 1993 are continuing education past the undergraduate level.  RS considered a 
standard model relating education, experience, and earnings (1974).  Their findings 
illustrate that the median (gross) earnings for college graduates (16 years of education) 
are roughly 60 percent higher than those of high school graduates (12 years of education), 
while high school dropouts earn about 32 percent less than individuals who have a high 
 17
school diploma (Rupert and Schweitzer 1996, p. 1107). In 1963, 42 percent of the 
fulltime workforce consisted of high school dropouts; by 1993, this figure had fallen to 
11 percent (Rupert and Schweitzer 1996, p. 1110).  The fraction of workers with only a 
high school diploma also declined over this period from 36 to 34 percent.  The share of 
the workforce holding a college degree rose from 7 to 18 percent, and the fraction with 
some post-graduate studies shot up from 3 to nearly 9 percent (Rupert and Schweitzer 
1996, p. 1110).  The change in measured experience fell by about fours years, from 24.1 
to 19.8.   This decline in labor market experience is the result of additional education 
requiring time out of the labor market, since RS measure experience as age minus years 
of education.  Even so, completing some college level courses does increase earnings by 
about 20 percent compared to people who only graduated from high school.  Completing 
college or graduate school increases the return to nearly 60 percent (Rupert and 
Schweitzer 1996, p. 1111).  In 1963, high school dropouts earned about 22 percent less 
than high school graduates; by 1993 they were earning about 32 percent less (Rupert and 
Schweitzer 1996, p. 1112).  RS also concentrate on the sex and race effects.  In 1963, 
blacks were paid roughly one-third less than whites (Rupert and Schweitzer 1996, p. 
1113).  By 1973, the gap had narrowed to about 20 percent, and by 1993, to about 16 
percent.  
Goetz and Freshwater (2001) discuss explanations of differences in economic 
growth rates.  They use a conditional convergence growth model which has introduced 
the possibility of incorporating a wide set of factors as determinants of growth.  They 
formulated a county-level empirical growth model that relates local per capita income 
growth to the social and institutional dimensions of a locality.  GF found that investment 
 18
                                                
in human capital; accessibility and adjacency to urban areas are more beneficial for 
economic growth.  Their empirical results indicate that, ceteris paribus, social and 
institutional variables explain some of the differences in convergence rates among 
counties in the United States.  For example, ethnic diversity is associated with faster rates 
of economic growth, higher levels of income inequality are associated with lower growth 
rates, and higher levels of social capital have a positive effect on economic growth rates 
(Goetz and Freshwater 2001, p.144).  The neoclassical growth model suggests that poorer 
countries and regions should grow faster than richer countries and regions, and poor 
countries and regions should eventually catch up based on the assumption of decreasing 
returns to capital (Goetz and Freshwater 2001, p.145).  Convergence in most cases is 
slow enough that the impact is far less than expected. Convergence is conditioned by 
various factors such as property rights, trade barriers, transaction costs, government 
policies, income inequality that impedes development, weak legal and business 
institutions, capital market imperfections and cultural differences (Yeager 1999).  This 
article contributes to the conceptual and empirical debate on the importance of income 
inequality, ethnic divisions, and social capital in stimulating local economic development 
in the United States.  The results support the argument that ethnic diversity in a locality is 
conducive to economic growth, because costs associated with spatial mismatch
2
 are 
lower in more ethnically diverse communities (Goetz and Freshwater 2001, p.146).  The 
estimated effect of income inequality on the growth process was found to be negative and 
significant for U.S. counties, even after accounting for the effect of all other explanatory 
variables.  Counties with a higher income inequality experience slower income growth.  
 
2
 The less likely a community will become more diverse within a county.  
 19
                                                
Their results expose that the traditional county level ordinary least squares model is 
misspecified due to the presence of spatial dependence
3
.  GF?s results confirm some of 
the major findings of the existing convergence literature, such as the negative impact on 
conditional convergence and a positive impact of education.  They also provide new 
evidence for some of the previously considered variables.   
Durden and Gaynor (1998) measure the costs, in reduced yearly earnings, of 
being black or Hispanic (Durden and Gaynor 1998, p.95).  DG wanted to see if earnings 
differentials by gender are greater than race or ethnic-based differentials, they then argue 
that gender specific policy tailoring might be needed.  They also demonstrate that if 
earnings differentials for black and Hispanic females exceed pure gender differentials, the 
extra gap is assumed to be the negative result of being neither white nor male.  DG?s 
model is based on standard human-capital theory, and ordinary least squares is used for 
the empirical model.  A significant difference in earnings exists between white males and 
all others, but when the earnings of black and Hispanic males are adjusted for differences 
in labor market characteristics the total earnings differentials disappear (Durden and 
Gaynor 1998, p.98).  This was not found to be true for white, black and Hispanic females. 
The total earnings difference for white or black females is nearly twice that for black 
males and Hispanic males are even greater.  Durden and Gaynor found that white male-
other group earnings gaps still exist, and that the cost of being black or Hispanic is far 
less than the cost of being female.   
 
 
 
3
 A U.S. Study by Rey and Montouri (1999) illustrated how although the regional convergence 
phenomenon has geographical dimensions, spatial effects have largely been ignored in the literature.   
 20
Public and Private School Education 
a. Public vs. Private Education       
In his article Sonstelie (1982) introduces how public schools in California are not 
free.  He implies that parents who send their children to public schools impose a cost on 
society as a whole, because scarce resources are employed in these schools.  From a 
parent?s perspective, public schools are free.  A family may have been taxed to support 
those schools, but it gains no additional cost if it sends its children to a public school nor 
is it relieved of any of its tax burden if it enrolls its children in a private school (Sonstelie 
1982, p. 794).  This discrepancy between the social and private costs of public schools 
can lead to an inefficient distribution of students between the public and private sectors.  
The welfare cost of the inefficiency is defined as follows. Let a family?s pubic school 
surplus be the maximum it is willing to pay to enroll its child in public school rather than 
in a private school of his or her choice (Sonstelie 1982, p.794).  The private schools vary 
in quality, and the tuition of each reflects its quality.  The alternatives are therefore a 
tuition-free education in the public sector, and the opportunity to buy an education of 
higher quality in the private sector.  Sonstelie analyzes the choice between these 
alternatives.  He views it as a derived utility, U(x,y), from two sources: the quality of its 
child?s education , y, and the income available to spend on all other goods, x (Sonstelie 
1982, p. 795).  He let q be the quality of the local public schools, and let w be the 
family?s income after taxes (Sonstelie 1982, p. 795).  Therefore, its utility is U (w,q).  
The family?s utility from opting for a private education is u* = U (w ? py*, y*), where y* 
represents the quality of the private school and p represents the price of education quality 
in the private sector (Sonstelie 1982, p. 796).  For that reason, it will choose a public 
 21
education for its child if and only if U (w,q) >u*.  Sonstelie establishes that free public 
schools give rise to a welfare cost, which is the difference between the cost of educating a 
child in public schools of quality, and the surplus the family derives from sending its 
child to public schools of quality rather than private schools of its choice (Sonstelie 1982, 
p. 796).  He believes parents will choose public schools if and only if their utility, which 
consists of the family?s income and quality of the school, exceeds the price and quality of 
the private school.  Parents? choice between a public and private sector education is based 
on the quality of the school.  Sonstelie found that welfare costs in California are 
substantial and are due mostly to lower unit costs in the private sector.  He establishes 
that families in California who are most apt to send their children to private schools 
might be expected to locate in communities where the quality of public schools is low.  
Figlio and Stone (2002) discuss whether local policies can improve the 
competitiveness of public schools in retaining students, particularly those students most 
likely to opt out- white, high-achieving students from well-educated, high-income 
families.  This paper examines the role of local policies and attributes, as they interact 
with individual student characteristics in determining the composition of public and 
private- school enrollments in particular communities (Figlio and Stone 2002, p.76).  
They talk about other studies that utilize data on student and family characteristics in 
eight New York metropolitan areas.  They argue that race plays the most important role.  
FS data came primarily from the restricted- access version of the National Educational 
Longitudinal Survey, conducted by the National Census of Education Statistics of the 
U.S. Department of Education (Figlio and Stone 2002, p.77).  Their model estimates the 
probability of student enrollment in a public or private school in the tenth grade using 
 22
individual student, family and community demographic and school characteristics, with a 
special emphasis on the interactions between individual and local policy-related variables 
(Figlio and Stone 2002, p.77).  FS adopt a standard set of family and demographic 
variables: variables reflecting gender, race, Hispanic ethnicity, the student?s family 
income, parental education, parental frequency of religious attendance, variables 
reflecting whether a male or female parent or guardian lives in the student?s home, an 
indicator for parental marital status, the number of siblings, fifteen religious 
denomination variables, the student?s eighth-grade mathematics test score, and nine 
Census-sub region dummies reflecting unobserved regional variations in relative 
probabilities of school- sector choice (Figlio and Stone 2002, p.78).  FS find that local 
violent crime rates, school districts, concentration, and student-teacher ratios in public 
schools, have a powerful role in determining the composition of public and private 
schools-especially in retaining those students most likely to opt out of public schools 
(Figlio and Stone 2002, p.78).   This article found that high-achieving students and 
students from higher income, better educated families, or families where at least one 
parent attends church regularly are more likely to attend a private school.  They also 
found that the more concentrated public schools are in a county, the more likely students 
are to select into private schools. 
Hamilton and Macauley (1991) discuss how published studies have generally 
used standardized test scores, or their year-to-year improvements, to measure inputs and 
outputs of education.  Yet it has long been recognized that these indices poorly represent 
the product demanded by parents.  HM specify an optimization model for the household 
facing a choice between public and private schools and derive an estimating equation that 
 23
enables them to make inferences about parameters of both education technology and 
demand.  The key variables discussed in the model are mean household income and its 
standard deviation within a school jurisdiction.  It was hypothesized that income 
dispersion plays a role in the technology for public (but not private) education. The 
standard deviation was included in the model to capture the nonlinear effect of (income-
proxies) student?s peer group on achievement.  Both income measures were found to be 
statistically significant (Hamilton and Macauley 1991, p.283).  This approach to school 
?output? bypasses a difficult measurement problem in investigating the technology and 
demand for education.  The demand for education takes the following form:  q
d
 = y?
1
p?
2
, 
where income and price elasticities are represented by ?
1 
and ?
2
, respectively (Hamilton 
and Macauley 1991, p.283).  The technology assumes slightly different forms for public 
and private education.  HM express it as a cost function.  For public schools, p = p
0 
?
?1
?
?2
, where ? is the mean income of public-school patrons and ? is the standard 
deviation of income (Hamilton and Macauley 1991, p.284).  The private school patron is 
represented by p= p
0 
y
?1
, where y is the income of the individual demander (Hamilton 
and Macauley 1991, p.284).  The model of household choice between public and private 
schools allows us to infer parameter values without having to measure achievement or 
other dimensions of school quality.  HM investigation also suggests that income and its 
standard deviation within a school jurisdiction are significant factors in the education 
production technology.  Some parameters do have large effects on the cost of producing 
education. 
 
  
 24
Another article that discusses the choice between different types of schools is 
Lankford and Wyckoff (1992). They use the history of public-private school choice to 
empirically investigate the determinants and consequences of public-private school 
choice.  The empirical results of this article focus on the choices made by individual 
families.  The model specifies the examples of school inputs, school orientation and 
environmental inputs and how these aspects may be pertinent for families selecting 
among school alternatives.  Other attributes of the model include school staffing levels 
and quality, quality of facilities, and academic standards and curriculums.  LW?s 
empirical model of how families choose among school alternatives is based upon the 
theoretical model and data that measure the factors hypothesized to affect school choice 
(Lankford and Wyckoff 1992, p.317).  They posit a rational model of choice: each family 
of a school-age child is assumed to evaluate its restricted set of mutually exclusive school 
options and to select the alternative that maximizes its utility (Lankford and Wyckoff 
1992, p.318).  They use a set of alternatives represented by J where U
mj
, j = 1, 2?J, is 
the utility of the mth household that would result if the jth school alternative were 
selected (Lankford and Wyckoff 1992, p.318). Alternative i is chosen if and only if U
m1
 > 
U
mk
 for all k ? i (Lankford and Wyckoff 1992, p.318).  The specification of U
mj 
is to be a 
function of school attributes the bundle of other goods and services consumed by the 
household and its tastes and preferences (Lankford and Wyckoff 1992, p.318).  LW 
implied econometric specification is a random utility model, where the probability that 
the mth household selects the ith school alternative is P
m1
 ? P( U
m1
 ? U
mk
, for all k? i) 
(Lankford and Wyckoff 1992, p.318).  LW felt that each attribute can affect how well a 
student achieves academic goals.   Average scores on standardized tests and college 
 25
attendance rates for a school were considered as two potentially important alternatives for 
such environmental factors.  However, test scores are important when deciding between 
public and private school education.  One important factor when sending a child to a 
private school should be the potential for the child to flourish intellectually or 
emotionally in public schools. 
Economic Factors that Affect Education 
a. Effects of dropout rates in High School 
Domazlicky (1996) measures the cost of high school non-completion in Southeast 
Missouri.  This article introduces the counties? average rate of high school non-
completion.  Domazlicky proposed that there are many significant economic and social 
costs associated with high school non-completion.  His economic model developed cost 
measurements of high school non-completion.  His model found that for every percentage 
point increase in a county?s high school non-completion rate, per capita income in the 
county falls by $52.  As a result, the high level of non-completion rates in Southeast 
Missouri led to a loss of over $400 million in total income in the counties that were 
studied.  
 Lillard and Decicca (1988) provide evidence that suggests students are more 
likely to drop out when they face higher standards.  This implies that individuals are more 
likely to drop out of high school in states with higher course graduation requirements 
(CGR).  LD?s data consist of aggregate data on dropout rates from 1980 through to 1994.  
In their empirical investigation they confront three potential estimation issues: omitted 
variables bias, the endogeneity of the policy variable of interest, CGR?s, and estimation 
of proper variances (Lillard and Decicca 1988, p.462).  Based on these issues the general 
 26
estimating equation is: Pr (dropout) = ?o + ?1 CGR + ?W + ?, where CGR is the course 
graduation requirement, W is the expected wage and ? is a person-specific residual term 
(Lillard and Decicca 1988, p.462). They found that in 1989 the average annual earnings 
of male dropouts between the ages of 18 and 24 were $8,046 (Lillard and Decicca 1988, 
p.463).  LD found that high school graduates earn about $11,741.  LD state that 
individual dropout students believe they receive one more year of earnings than 
graduates, however, their extra earnings will be wiped out in about two years.  They 
establish that from the point where the student drops out of school, they lose an average 
of $3,700 per year (Lillard and Decicca 1988, p.464).  LD calculates the potential losses 
dropouts might suffer when graduation requirements are raised.  Their main concern is to 
investigate whether changes in graduation requirements affect decisions to remain in 
school.  Therefore, based on their evidence, it is implied that more people will drop out of 
high school if requirements are raised.   
 Betts and Grogger (2003) emphasize their empirical work on the effects of 
higher grading standards.  They established that many economists focus on readily 
measurable inputs, such as class size.  Relatively little has focused on the incentives that 
schools create for their students.  A common finding is that higher standards may help 
some students at the expense of others (Betts and Grogger 2003, p.343).  For example, 
higher standards may lead more motivated students to increase effort; they may cause 
others to give up as the standard moves beyond their reach.  BG expands on existing 
empirical work in two ways: First, they analyze the effect of schools? grading standards 
on a number of educational outcomes (Betts and Grogger 2003, p.343).  They estimate 
the effects of higher standards not only on test scores but also on high school graduation, 
 27
college attendance, and entry-level earnings (Betts and Grogger 2003, p.344).  Secondly, 
they consider the consequences of higher standards.  BG first use regression methods to 
estimate effects at different levels in achievement distribution.  Then they categorize the 
sample by race/ethnicity to decide whether higher standards have adverse consequences 
for minorities.  BG use data from the Cohort of High School and Beyond Survey, which 
provides a sample of approximately 15,000 students from 1000 different schools (Betts 
and Grogger 2003, p.344).  They use regression methods to construct grading standards 
for each school.  The school?s grading standard is the achievement level needed for 
students to receive any given grade (Betts and Grogger 2003, p.334).  This measures how 
strenuously the school grades its students, relative to an objective measure of student 
achievement obtained from standardized tests.  For example, one school generally gives 
B?s to students who score in the 75
th
 percentile on the nationwide test, whereas a second 
school generally gives such students C?s, then the later school has higher grading 
standards (Betts and Grogger 2003, p.334).  BG construct grading standards using 
regression methods which demonstrate the impact on educational outcomes.  Their model 
followed this format: y
ij 
= ??
j
 +  X
ij
?
  
+ Z
j
? + u
ij
, where y
ij
 is the outcome measure for 
the ith student in the jth school, X
ij
 is a vector of student background characteristics, Z
j
 is 
a vector of school characteristics, and u
ij
 is a disturbance term (Betts and Grogger 2003, 
p.336).  The terms ?,?,?,are the parameters to be estimated, where ? gives the effect of a 
unit change in the grading standards on the educational outcome (Betts and Grogger 
2003, p.336).  BG results explain that higher grading standards do raise test scores, and 
they have no significant effect on educational attainment.  They believe that this mystery 
could be due to the fact that grading standards have their greatest effect among the 
 28
students who are the most likely to graduate (Betts and Grogger 2003, p.339).  Therefore, 
one might imagine that large effects on test scores could lead to increases in graduation 
rates.  However, the effect of grading standards at the bottom quartile is relatively small, 
which may explain why they have little effect on graduation rates.   
b. Effects of Educational Spending 
 Weiss (1970) examines the effect of schooling and learning on the level of 
workers? earnings.  He obtained his data from Equality of Educational Opportunity, 
which measures the effect of educational attainment and various other personal 
characteristics on the earnings of those with twelve or fewer years of schooling (Weiss 
1970, p.150).  Weiss discusses the specification of the earnings functions and the 
significant relationship that exists between an individual?s scholastic achievement and 
their earnings.  Weiss uses an earnings function to test his hypotheses. He estimates:  
Y= B
0
 + B
1
M + B
2
A + B
3
Mw + B
4
W + B
5
V + B
0E
, where the Y represents annual 
earnings in 1959, M is whether or not the individual is married, A represents age, Mw 
represents whether the individual?s work place in a metropolitan area or outside a 
metropolitan area, V represents whether the individual is a veteran or not, and B 
symbolizes estimated achievement in school (Weiss 1970, p.150).  Weiss uses the 
achievement variable in his model because he believes it is a better measure of skills or 
productivity than merely the number of years in school.  It allows him to take account of 
well-documented variations in school achievement in evaluating the effect of education 
upon earnings.  Weiss uses two different types of earnings functions.  He estimates white 
and blacks earnings separately.  He compares his estimates and establishes that the 
achievement of blacks is generally much lower than that of whites at every grade level 
 29
(Weiss 1970, p.154).   However, it is important to understand that this low achievement 
level is positively related to the generally low earnings of blacks.  Weiss finds that from 
an average year of school, a black gains only 0.5 to 0.8 units of achievement (Weiss 
1970, p.155).  He also found that the employment variable, which is estimated in the 
black earnings function, accounts for all the explained variance for blacks.  This aspect 
establishes that lower black levels of earnings cannot be completely explained by their 
lower achievement.  Two of Weiss?s hypotheses tested in this study (a regular and 
significant relationship exists between scholastic achievement and earnings and 
achievement explained more of the variance in earnings than does the number of years in 
school) were only substantiated for whites (Weiss 1970, p.159).  However, for blacks 
only one out of four age groups considered demonstrated that years of schooling or 
achievement is positively related to earnings.  Not surprisingly, a black has lower 
earnings than a white with the same years of schooling.  This suggests that the solution to 
black poverty is outside the classroom 
 Ribich and Murphy (1975) discuss the effectiveness of schools producing 
learning and other beneficial outcomes.  They find that extra educational spending does 
lead to greater lifetime income, but that the income gain is less than the amount of the 
extra spending required to induce it.  RM decide to estimate their model using the human-
capital approach and the production function approach.  Through their models they try to 
explain the monetary earnings of individuals.  They propose that individuals? earnings are 
in large part determined by his or her ability, aptitude, and achievement, the time spent in 
formal or informal educational training, the cost and quality of that schooling and 
training, the socioeconomic status of the parents, and the socioeconomic status of 
 30
?others? closely associated with during the ?formative years? (Ribich and Murphy 1975, 
p.58).  RM considers these variables to be the ?supply side? variables.  By testing their 
models they hope to explore the possibility that this lack of influence on test score 
performance does not necessarily mean that increased expenditure yields no long-run 
benefits whatsoever (Ribich and Murphy 1975, p.59).  RM demonstrate that expenditures 
per student are positively related to test scores in the ninth grade, years of schooling 
completed, and lifetime income (Ribich and Murphy 1975, p.64).  Therefore, test scores 
are positively related to years of schooling completed and to lifetime income.  RM?s 
study establishes that more spending on educational attainment does give rise to 
increased earnings mainly through the indirect effect of increased educational attainment, 
except in those subgroups where sample bias and the range of spending variability is 
small (Ribich and Murphy 1975, p.74).  Their results suggest that the emphasis of past 
studies on the test scores in the evaluation of schooling changes is inappropriate.  Even 
when the test score changes associated with a schooling change are negative, the 
measured long-term effects on important aspects such as school continuance and on 
lifetime income may be sufficient to validate undertaking the change anyway.    
Bowels and Levin (1968) discuss how per-pupil expenditures and school facilities 
show very little relation to student achievement levels, and the effect of a student?s peers 
on his achievement level is more important than any other school influence.  Their paper 
scrutinizes the data and the statistical analysis on which these findings are based.  It is 
suggested that because of poor measurement of school resources, inadequate control of 
social background, and inappropriate statistical techniques used in the presence of 
interdependence among the independent variables, many of the findings of the report are 
 31
not supported (Bowels and Levin 1968, p.4).  For example, the survey revealed a 
significant amount of segregation in the North as well as in the South, relatively minor 
differences in the measured characteristics of schools attended by different racial and 
ethnic groups, and very great differences in the achievement levels of racial and ethnic 
groups throughout the county (Bowels and Levin 1968, p.4).  BL confine their attention 
to that section of the study that examines the relation of school and student characteristics 
to scholastic achievement.  They found that most of the conventional measures of school 
resources ? per-student instructional expenditures, facilities, pupil-teacher ratio, and 
curriculum ? accounted for very little of the variance in achievement scores of students.  
BL use linear regression to relate achievement to student background and school 
characteristics is the linear regression model (Bowels and Levin 1968, p.6).  Their model 
illustrates that school inputs and the social background and attitude characteristics of the 
students and their peers can be used to predict current school outcomes, as measured by 
achievement scores.  This model implies that a student?s achievement level is merely the 
sum of the independent effects of each school resource and background variable, plus a 
constant (Bowels and Levin 1968, p.9).  BL attempted to show from a previous study that 
both the measurement of the student was inadequate, and that the statistical techniques 
were inappropriate.  The study they examined stimulated a great deal of thought and new 
research efforts to uncover the largely unknown and complex relationships among family, 
school, and community influences on one hand, and educational outcomes on the other 
(Bowels and Levin 1968, p.10).  BL study did not provide answers to their questions; it 
did bring them closer to being able to use large scale research efforts as a basis for 
making intelligent policy decisions for our schools.   
 32
                                                
Rhine?s (1989) main purpose in her study is to investigate whether or not 
performance differences exist between students in a mandated and non-mandated
4
 
environment.  For several decades, economic educators have attempted to identify factors 
that influence a student?s performance in economics.  Several Studies suggest that 
learning economics, as measured by performance, is enhanced by various attributes such 
as maturity, ability, effort, and curriculum preferences (Rhine 1989, p. 231). 
Socioeconomic factors such as family income and parental occupation also appear to 
have an important impact on a student?s ability to understand economics.  The evidence 
in this article suggests that being male may have a positive effect on the initial level of 
performance (for-example, a pre-test) but little, if any, effect on final performance (a 
post-test score).  Student performance is analyzed through a descriptive model of learning 
which is defined as L= f [HC, Q, E, M] where L= learning, HC = vector of human capital 
characteristics, Q = vector of teacher quality characteristics, E = environment factor(s), 
and M = state mandate (Rhine 1989, p. 232).  According to the model, learning is 
dependent upon various human capital characteristics, the quality of instruction, the 
learning environment, and the presence of a state mandate.  For example, some human 
capital traits are aptitude, and personal characteristics include sex and race.   The overall 
number of years of experience of the teacher is included in the vector Q. Environmental 
factors (E) include the size and location of the school district. The hypothesis tests 
whether or not performance differences exist between students in mandated and non-
mandated environment.  Rhine emphasizes that when estimating a student?s performance 
it is important to control for a student?s initial level of aptitude.  Students with a higher 
 
4
 Definition:  To make mandatory, as by law; decree or require. 
 33
aptitude for economics are expected to perform better than students with a lower aptitude 
(Rhine 1989, p. 233).  Since student aptitude is unobservable, a two-stage least square 
(2SLS) technique is used to avoid simultaneous equations bias.  Using pre-test and post-
test performance as the students? aptitude and performance measures, a two step system 
is specified as Pre test = ?
0
 + ?
1 
X + ?
1
 and Performance = ?
0
 + ?
1
X' + ?
2
PRETEST + 
?
3
M + ?
2
, the vectors of human capital, teacher quality, and environmental characteristics 
are represented by X and X' (Rhine 1989, p. 233).  Rhine?s results suggest that structural 
differences in performance do exist between students in a mandated and non-mandated 
environment.   
Hanushek (2004) states that the starting point of his debate has been an empirical 
finding-that resource for schools do not appear closely related to student performance.  
He establishes that data on school resources and student performance have been available 
for a substantial period of time.  He found that there has been a dramatic increase in the 
resources for schools between 1960 and 2000.  Pupil-teacher ratios have fallen by a third; 
the proportion of teachers with a master?s degree has more than doubled; average teacher 
experience has grown to new highs; and real spending per pupil has more than tripled 
over the 1960-2000 period (Hanushek 2004, p. 157).  These patterns in resources would 
be expected to lead to improvements in student performance, but Hanushek illustrates 
that this is not the case.  The data used do suggest that some aspects of family have 
changed in ways that would generally be thought to be unfavorable to student learning: 
an increase in family poverty (until the early 1990s), a larger proportion of children from 
single parent families, and more children from non-English speaking families (Hanushek 
2004, p.158).  The summary of the basic evidence is straightforward.  Significant 
 34
differences in the performance of schools exist.  These differences, however, are not 
easily captured by simple measures of resources.  A surprisingly large proportion of these 
studies even suggest that added funds actually lower achievement (Hanushek 2004, 
p.159).  It seems hard to imagine that adding additional resources actually depressed 
student performance very frequently.  The model of student performance was considered 
as: O
i 
= F
i
T
i
E
i , 
where O is the student performance, F is family input, T is teacher input, 
and E is school expenditure.  This relationship would in fact recognize the dual role 
played by teachers.  The impact of resources is complicated ? involving interactions with 
various inputs that are not observed or are not understood.  Hanushek illustrates in his 
calculations that teacher quality essentially determines the efficiency with which 
resources are converted into student achievement.  
 c. Differences in student performance between race 
 Hanushek (2001) discusses how human-capital differences have provided a 
common explanation for racial earnings differences, although for data reasons most 
attention has focused just on differences in school quantity and has left out quality 
considerations.  Trends in the cognitive achievement scores compiled by the National 
Assessment of Educational Progress (NAEP) provide consistent quality data since 
roughly 1970 for a representative sample of students across mathematics, reading, and 
science (Hanushek 2001, p. 24). Hanushek suggests that three major systematic factors 
offer clear possibilities for explaining the changes in black-white performance on the 
NAEP tests.  Over the past three decades two major governmental interventions with 
significant racial dimensions have had enormous effects on school operations: legal 
actions to promote school desegregation and legislative and legal actions to change the 
 35
level and distribution of school funding (Hanushek 2001, p. 24).  David Grissmer et al. 
(1998) point to added resources focused on blacks, a result of spending increases and 
more equalized spending across districts.  Michael Cook and William Evans (2000), on 
the other hand, find that narrowing of NAEP scores between blacks and whites cannot be 
any simple reflection of school funding levels, since three-quarters of the gap lies within 
schools.  Thus, if resources were a prime factor behind the narrowing of test-score 
differences by race, it must be the case either that minorities are more sensitive to 
resource differences than white students or that resources are narrowly targeted to blacks 
within schools (Hanushek 2001, p. 24).  Hanushek believes that simple cross-state 
educational production-function estimation tends to confuse the different state policy 
environments with the effects of resources, but test-score-gain models reduce this 
problem.  Hanushek found that within-state variations in spending could have an impact 
if more equalized spending tends to favor black students.  Hanushek?s estimates have 
considered whether any of the governmental or family factors individually could explain 
the magnitude and pattern of black-white achievement gaps.  Neither the level nor the 
distribution of school spending appears to provide much explanation for gaps.  School 
spending levels show little consistent impact with any indication of differential impact on 
blacks being small. Direct analyses of the effects of spending equalization on 
performance similarly show little impact.  On the other hand, governmental intervention 
through integration programs appears potentially more important.  The pattern of 
integration and preliminary estimates of the magnitude of effects suggest that this by 
itself could explain both the narrowing and the subsequent leveling off of gains.  
Hanushek finds there are two facets to the estimated impact of both integration and 
 36
parental education.  The evidence suggests that blacks are more sensitive than whites to 
each of these factors (Hanushek 2001, p. 26).  Furthermore, the patterns of change in the 
factors (more integrated schools and improvements in parental education) have both 
favored blacks.   
Economics Factors that Effect Per Capita Income  
a. Education and Per Capita Income 
 Patrinos (1997) examines the earnings gap between indigenous and non 
indigenous people in Guatemala.  There is a gap in earnings between the two groups.  He 
demonstrates that the gap is due to a difference in characteristics of ethnic groups that are 
also segregated by occupation and industry. They have differential access to resources, 
power and regional centers, which includes metropolitan Guatemala.  The Guatemalan 
household survey used in this article identifies indigenous people the language spoken.  
Ethnic group affiliation is determined on the basis of language. Ethnicity has had a 
dramatic impact on schooling, earnings, and returns to schooling.  It has been shown that 
in the United States, for example, Hispanics do better or worse depending on where in 
Latin America they are from (Patrinos 1997, p. 810).  The socioeconomic performance of 
workers not only depends on the skills of their own parents but also on the abilities and 
skills of the ethnic groups.  The question of ethnicity is examined in terms of schooling 
attainment and the returns to schooling.  Patrinos tests the hypothesis that higher levels of 
schooling correspond to higher returns to schooling.  In addition to estimating the returns 
to schooling as derived from earnings functions, the results are also used to estimate the 
extent of ethnic group discrimination (Patrinos 1997, p. 812).  The empirical findings 
demonstrate a high return to schooling.  The data used include both household-level 
 37
information and individual-level information.  The information was collected in urban 
areas.  The model consists of two equations, or earning functions, which are fitted for 
employed members of the economically dominant group (nonindigenous) and employed 
members of the marginal group (indigenous): LnY
n 
= b
n
X
n
 + u
n 
and LnY
i 
= b
i
 X
i 
+ u
i
, 
where subscripts n and i represent nonindigenous and indigenous workers; Y symbolizes 
labor market earnings; X represents measured productivity-determining characteristics of 
the workers, such as education and experience (Patinos 1997, p.813).  The results reveal 
that higher levels of schooling correspond to higher returns to schooling within each 
ethnic group.  Schooling levels are lower for indigenous groups, especially for the 
Quiche, but the rates of return to schooling are higher for indigenous groups with high 
levels of schooling than for less schooled groups.   
Greene and Hoffnar (1994) report three principal findings in their article.  First 
the authors found that both central city and suburban Hispanics earn significantly less 
than white male residents of the suburbs (Greene and Hoffnar 1994, p.127).  Secondly, 
the results demonstrate that while central city residence lowers the earnings of both white 
and Hispanic men, the impact is more pronounced for Hispanics.  Third, this study did 
not find any evidence that suburban Hispanics learn more than central city Hispanics.  
GH believe that these results may imply that ?race rather than space? may play a more 
important role in contributing to the white-Hispanic earnings differential among males.  
They debate the impact that residential location has on earnings determination and 
whether the effect of residential location on earnings varies across different demographic 
groups.  GH use a standard approach to assess the impact of residential location.  They 
use a log earnings equation: Y
i 
= X
i
? + ?
i1
.  The X
i 
is a vector of characteristics believed 
 38
to influence earnings, ? is a vector of parameters to be estimated and ? is a stochastic 
error term (Greene and Hoffnar 1994, p.148).  However, a sample selection problem did 
occur, because earnings information was only available for those individuals who 
participate in the labor force (Greene and Hoffnar 1994, p 148).  Therefore, GH 
considered an indicator variable, S
i
, defined in such a manner that individual earnings are 
only observed if S
i
 >0 (Greene and Hoffnar 1994, p 148).  Let S
i 
be modeled as a binary 
process: S
i
 = Z
i 
? + ?
i2, 
where Zi and ? respectively, represent the determinants of labor-
force participation and the parameters to be estimated (Greene and Hoffnar 1994, p 148). 
Averett and Burton (1996) examine gender differences in the decision of whether 
or not to attend college.  These authors demonstrate how the decade of the 1980s saw a 
sharp rise in the already well-documented wage premium available to college graduates.  
This paper illustrates the role of the college wage premium and how it affects the 
probability of attending college and identifies any gender based variations.  AB 
demonstrate how the probability of a wage gap could be the result of differences in how 
men and women perceive and value the pecuniary and non-pecuniary rewards of higher 
education.  The percentage of female high school graduates who choose to attend at least 
some college rose steadily over the period of investigation while the percentage of males 
making this same decision has remained largely constant. AB show how college 
attendance rates for males and females have converged.  They suggest that it is possible 
that there have been significant changes in the value women place on the indirect benefits 
associated with a college education or that the absolute value of these benefits has 
increased.  AB also believe that it is possible that women are now better able to translate 
educational attainment into greater earnings and that this has motivated the increase in 
 39
college attendance.  A third explanation is that the costs associated with educational 
attainment have risen more slowly for women than for men.  Finally, AB found that the 
relative long-term shift in the college attendance rates of men and women may be 
associated with the way each group has come to value the earnings premium associated 
with college attendance.  Their primary aim in this article is to capture and analyze 
gender variations in schooling decisions using a human capital model.   
The majority of their data were obtained from the National Longitudinal Survey of Youth 
(Averett and Burton 1996, p.38).  They are primarily interested in how expected future 
earnings influence the decision to attend college.  They drew their initial sample from the 
1981 interview year when the respondents were aged 16 to 23 years (Averett and Burton 
1996, p. 39).  AB?s dependent variable equals one if the respondent attended some 
college and zero otherwise (Averett and Burton 1996, p. 39).  AB based their assumption 
of individual utility maximization, the probabilities that an individual i will choose a high 
school education, H or a college education, C, was expressed as: Pr (H) = Pr (U
IH 
? U
IC
), 
Pr (C) = Pr (U
IH
 ? U
IC
) where U is the present value of the anticipated stream of utility 
associated with some particular level of education (Averett and Burton 1996, p. 40).  This 
can be expressed as a ?structural probit? model.  The model is: C = ??Z + ?(y
c ? 
y
H
) ? ?, 
I= 1 if C>0 and I = 0 otherwise.  The decision to attend college serves as a dependent 
variable in this equation.  This is modeled as a function of family background 
characteristics, by vector Z, and the earnings differentials between high school and 
college graduates, (y
c
 ? y
H
) (Averett and Burton 1996, p.40).  The family background 
variables capture differences in financial constraints (number of siblings, race and 
ethnicity) and tastes (parental occupation, access to library cards, magazines and 
 40
newspapers when growing up, religion and region of residence in adolescence) (Averett 
and Burton 1996, p.40).  Their analysis revealed that those who attend college have better 
educated parents and fewer and more educated siblings (Averett and Burton 1996, p.41).  
The authors established that the behavior is consistent with the comparative advantage 
hypothesis for both the men and women who choose not to attend college.   
Tamura (2001) considered whether human capital convergence could occur in a 
world in which poor parents, with lower levels of human capital, must finance their 
children?s accumulation of human capital out of their own resources.  If teacher quality is 
more important for human capital formation than individual instruction, poor families 
will converge to the level of rich families.  Tamuras considered the role of convergence 
in formal schooling in the convergence of incomes per capita.  The model shows that if 
teacher quality is more important for human capital accumulation than individual 
instruction, human capital convergence will occur between school districts.  Tamura 
discussed that because wealthy parents can afford to purchase more and higher quality 
school inputs, it is possible that children from poorer families are destined to lag their 
wealthier counterparts in human capital accumulation.  Tamura found that the 
productivity of formal schooling in producing human capital depends on two inputs: 
quality and class size (Tamura 2001, p.110).  Tamura found that human capital 
convergence across school districts will occur if teacher quality is relatively more 
important than class size in the production of human capital (Tamura 2001, p.111).  The 
model proposes that an individual?s life consists of two periods: youth and adulthood.  
 41
                                                
Tamura felt that youth spend their time acquiring human capital
5
.  A parent cares about 
his or her consumption as an adult and the human capital of his or her child (Tamura 
2001, p.112).  These preferences are demonstrated in this model.  In Tamura?s model all 
parents within a school district have identical preferences and identical human capital 
levels.  He explains that teachers can be hired from any district and they can teach in a 
district different from the one that they reside in (Tamura 2001, p.113).  Therefore, only 
two school districts were considered in Tamura?s model.  He let Q
iT 
be the average 
teacher human capital in district i relative to average parental human capital in i, that is, 
Q
iT
 = average human capital of school district i teachers divided by average human 
capital of school district i parent, which is represented by: E {?
iT
}/ E {h
iT
 (Tamura 2001, 
p.115).  He also let C
iT
 be the class size in school district i, that is, C
iT 
= number of 
students in school districts i divided by the number of teachers in the school district i 
(Tamura 2001, p.115).  He uses Q
IT 
and C
IT
 to explain the next generations? level of 
human capital, using a Cobb-Douglas specification.  The model shows that human capital 
convergence between poor, low-human capital districts is possible even when the quality 
of education that children receive is determined solely by the parents? resources.   
All studies recognized in the literature review demonstrate the effects of 
education on earnings in some form.  For example, Ferguson states how education does 
affect per capita income in certain counties of the state of Alabama. His model uses some 
of the same variables as the model in this paper.  The comparison of variables is strongly 
related and he demonstrates the significance of educational expenditure as well. On the 
other hand, Hanushek discusses the importance of public education and how it is more 
 
5
 Human capital is a set of skills which an employee acquires on the job, through training and experience, 
and which will increase that employee's value in the marketplace. 
 42
beneficial than private schooling.  Nevertheless, Figlio and Stone both illustrate the 
importance of private education and how it is more beneficial to children than public 
education.  Theses studies examine the advantages and disadvantages of public and 
private education and how education is beneficial to the community as a whole.  Each 
study emphasizes the importance of education not only in Alabama but worldwide.
 
 43
CHAPTER III 
 
MODEL AND VARIABLE ANALYSIS 
 
 Three equations are examined to demonstrate the importance of student 
performance on school quality, the factors that affect parental decisions to send their 
child to a private school?  How does education affect annual income?  Does going to a 
private school affect annual income?   
Income distribution has been a topic of persistent interest in economics.  In recent 
years, motivated by the availability of new data sets and advances in the theory of 
economic growth and development, there has been renewed interest in understanding the 
determinants and dynamics of income distribution.  The literature emphasizes education 
as one of the major factors affecting the degree of income inequality.  Some studies show 
that with more education people will have higher wages.  It would seem to follow 
naturally that if more individuals are educated, average income should rise.  For instance, 
the human capital income distribution model that has been demonstrated in previous 
works implies that the distribution of earnings (or income) is determined by the level and 
distribution of schooling across the population (Durden and Gaynor 1998, p.95).  
However, it has been found that for a given distribution of education, an increase in 
average schooling has an ambiguous effect on income distribution (Durden and Gaynor 
1998, p.95).  
 
 44
I will be estimating equations modeling student performance, income, and school 
choice. Estimation of performance and income will be done using a production function. 
The school performance model is the following:  
1.  Student Performance Model Description:  Linear model 
  
Conceptual Model: 
   Performance = f (Private, South, Z
1
) 
 The student performance conceptual model follows the existing literature 
including relevant exogenous variables, where Z
1
 is the vector of variables (assumed to 
be exogenous) that influence student performance.  The dependent variable is the change 
in test scores between two years, 1992 and 1988.  Many factors affect student 
performance; family effects, socio-economic effects, school characteristics, and personal 
attributes are considered to be the most important categories of factors to include.  The 
empirical student performance model is: 
Empirical Model: 
  POSIM= ?
0 
+ ?
1 
CATH + ?
2 
OTHREL+ ?
3
SUBUR + ?
4
RURAL + ?
5
ASIANPI 
+?
6
AMERIN + ?
7
HISPAN + ?
8
BLACK + ?
9
PCTLUN + ?
10
SALHIGh + ?
11
STRATIO + 
?
12
MTHS + ?
13
MTHS0 + ?
14
TENROLL + ?
15
THIRTF + ?
16
HUND + ?
17
TEN + 
?
18
PSOUTH + ?
19
PNORTH + ?
20
MALE + ?
21-69
DVSTATES 
 
Variables: 
Student Performance Variables: (Dependent Variable) 
1. Test 88:  Average 8
th
 grade test scores for Reading, Mathematics,     
Sciences, History, and Social Sciences for 1988. 
 
2. Test 92:  Averages 12
th
 grade test scores for Reading, Mathematics, 
Sciences, History, and Social Sciences for 1992. 
  
The dependent variable will be the differences in the test scores:  Var. Test         
92- Var. Test 88 
 
 
 
 45
Family Effects Variables:   
 
1.  MOMSCH:  Education of respondent?s mother 
(MTHS (more than high school), MLTHS (less than high school) 
 
 
2.  FATHSCH:  Education of respondent?s father 
(MTHS (more than high school), LTHS (less than high school) 
 
Family Effects Variables Cont: 
 
3. FAMINC:  Household income of the respondent 
 (TEN (less than 10,000 dollars  per year, THIRTF between 10,000 and 
34,999 dollars per year, and HUND is between 35,000 and 99,999 dollars 
per year) 
 
Socio-Economics Variables: 
 
1. SCHTYPE:  Dummy variable for Private, Catholic, Private Non-
religious, Non-accredited, public schools as the base category. 
(PUB, CATH, PRIVNON, PRIV, NACRED) 
 
2. DVSTATES: Dummy variables for each state.           
 (DVAL, DVGA, DVMS, etc?) 
 
      3.   URBN3: The schools location in an environmental characteristic 
            (URBAN, SUBUR, RURAL) 
 
School Quality Variables: 
 
1. TENROLL: Total student enrollment of the school. 
 
2. STRATIO:  Average student-teacher ratio for the school.  
 
3. SALHIGH: Represents the highest salary paid to teachers at the 
school.  
 
4. PCTLUN:  The Percent of students at the school on free 
 or price reduced school lunch. 
    
 
 
 
 
 
 
 46
Personal Attributes Variables:   
 
          1.  RACE:  Represents the Race of the student. 
                               (ASIANIL, AMERIN, WHITE, BLACK, HISPAN)     
 
 2. SEX:  The gender of the respondent/student.  
     (MALE or FEMALE)  
 
The school choice model is the following: 
 
2.  The School choice Model Description: Multinomial Logit  
 
Conceptual Model: 
     Private = g (Z
2
, South) 
The school choice conceptual model demonstrates the important factors that 
influence a parent to send their child to a private school.  The relevant variables follow 
the literature on the important factors that affect private school attendance, where the 
vector Z
2 
is the vector of the variables (assumed to be exogenous) that influence school 
choice.  The exogenous variables include family effects, school characteristics, personal 
attributes and community effects. The dependent variable is a categorical variable 
including all five school types.  The individual variables and the empirical model is: 
Empirical Model: 
ORDER = ?
0 
+ ?
1
FEMALE + ?
2
ASIANPI + ?
3
AMERIN + ?
4
HISPAN + ?
5
BLACK + 
?
6
URBAN + ?
7
SUBUR + ?
8
NORTHE + ?
9
MIDW + ?
10
SOUTH + ?
11
STRATIO + 
?
12
MTHS + ?
13
MLTHS + ?
14
TEN + ?
15
THIRTF + ?
16
HUND 
 
Variables: 
Private School Attendance: (Dependent Variable) 
 
  1.  ORDER:  School type  
     (PUBLIC, CATH, OTHREL, PRIV, NACRED) 
 
Family Effects Variables: 
 
1. MOMSCH:  Represents the respondent?s mother?s education.  
    (LTHS, MTHS0) 
 
 47
Family Effects Variables Cont.: 
 
2. FATHSCH:  Represents the respondent?s father?s education. 
(LTHS, MTHS) 
 
3. FAMINC:  Household income of the respondent. 
(TEN (less than 10,000 dollars  per year, THIRTF between 10,000  
and 34,999 dollars per year, and HUND is between 35,000 and 99,999 
dollars per year) 
 
School Characteristics:  
 
1. STRATIO:  Average student-teacher ratio for the school.  
 
2.  Test 92:  Average of 12
th
 grade test scores in tests of Reading, 
Mathematics, Sciences,History, and Social Sciences for year 1992. 
 
  3.  SALHIGH:  Represents the Highest salary paid to teachers at the 
school.   
Personal Attributes:   
 
          1.  RACE:  Represents the Race of the student. 
                              (ASIANPI, AMERIN, WHITE, BLACK, HISPAN)  
   
             2.  SEX:  The gender of the respondent/student. (MALE or FEMALE) 
 
Community Effects: 
 
1. URBN3: Represents the schools location as an environmental 
characteristic.   (URBAN, SUBUR, RURAL) 
    
2.   REGON:  Represents the schools geographic location.  
     (NORTHE, MIDW, SOUTH, WEST) 
 
The earnings model is as follows: 
3.  Earnings Model Description: linear model. 
Conceptual Model:   
 Income = h (Private, Performance, South, Z
3
) 
The earnings conceptual model determines the effect of standardized test scores 
and school choice on income, where Z
3
 is the vector of variables (assumed to be 
exogenous) that influence the student?s income at age 25.  These variables represent the 
 48
best factors that influence personal income.  The list of relevant variables and empirical 
model is as follows: 
Empirical Model: 
AINCOME = ?
0
 + ?
1
MALE + ?
2
CATH + ?
3
OTHREL + ?
4
ASIANPI + ?
5
AMERIN +  
?
6
HISPAN + ?
7
BLACK + ?
8
RACECUR + ?
9
NORTHE + ?
10
MIDW + ?
11
SOUTH + 
?
12
GRDDEG + ?
13
UNDDEG + ?
14
TEST 92 + ?
15
PNORTH + ?
16
PSOUTH 
 
Variables: 
Income Variable: (Dependent Variable) 
            
   1.  AINCOME:  Respondent?s annual income in 1999. 
     (divided by a thousand) 
 
Student Performance Variable: 
 
1. UNDGPA:  Respondents undergraduate GPA.     
 
2.  TEST 92 ? Represents students average test scores for the twelfth      
grade in subjects, reading, mathematics, science, 
History/citizenship/geography 
 
School Type Variables: 
 
        1.  SCHTYPE:  Represents Private, Catholic,  
       Private Non-Religious, Non-Accredited, and Public Schools. 
       (PUB, CATH, OTHREL, PRIV, NACRED)  
 
             2. URBN3: Represents schools environmental characteristics.   
         (URBAN, SURBUR, RURAL) 
 
Personal Characteristics Variables: 
 
1. RACE:  Represents the race of the respondent 
                               (ASIANIL, AMERIN, WHITE, BLACK, HISPAN) 
 
2.  SEX:  Represents the sex of the respondent 
   (MALE or FEMALE) 
     
Education Variable:  
 
  1.  RHDEG: Represents the respondents? highest degree earned. 
                              (NODEG, UNDDEG, GRDDEG)   
 
 49
Location Variables: 
 
1. RACECUR:  Represents the percent of people in their  
     current neighborhood who were of the same race.   
 
2. REGION:  Represents the schools geographic location.  
      (NORTHE, MIDW, SOUTH, WEST) 
 
Equations 1-3 make plain the model?s recursive structure.  It is also clear that 
South affects a student?s earning capacity directly via income equation and indirectly via 
student performance and school choice equations.  Specifically, taking the derivative of 
the income equation with respect to South holding all other things constant yields: 
?Inc/?South=(?Inc/?South)|  
dperformance = dchoice = 0 
+
 
 
(?Inc/?Choice)(?Choice/?South) + 
(?Inc/?Peformance)(?Performance/?Choice)(?Choice/?South) 
Or,  
4.    ?Inc/?South = (?h/?South) + (?h/?Choice)(?f/?South) + 
(?h/?Performance)(?g/?Choice)(?f/?South)    
The first term in equation 4 represents the direct effect of South on student earnings, the  
second and third terms represent the indirect effects.  The indirect effects depend on  
whether the South effects estimated in equations 1-3 are significant and, if so, their signs 
and magnitudes are determined.  In particular, if South in equation 1 is not significant, 
indirect effects vanish and equation 4 reduces to ?Inc/?South = (?h/?South). 
        
 
 50
Dataset 
 The data set used for estimation comes from the National Educational 
Longitudinal Study of 1988, 1992 and 2000 (NELS 88, 92, 00).  In these studies, a 
nationally representative sample of eighth graders was first surveyed in the spring of 
1988.  Later, a sample of these respondents was resurveyed through four follow-ups in 
1990, 1992, 1994 and 2000 respectively.  This dataset provides a rich amount of 
information by surveying not only the students, but also their parents, teachers, and the 
school administrators at the high school level. 
 Relevant data are used from the base year, the second follow-up and the fourth 
follow-up surveys.  Student performance at the high school level is measured by the 
average of the standardized NELS test scores (expressed as a percentile rank) at the 
twelfth grade minus and the scores in the eighth grade, to capture change in performance 
(POSIM).  In the dataset, there are test scores on four subjects: reading, mathematics, 
science and history/citizenship/geography.  Standardized, the test scores at the four 
subjects have almost the same means and standard deviation.  Consequently, POSIM 
takes the average of the available standardized test scores on the four subjects for each 
student to prevent collinearity and to minimize missing data problems.  When fewer than 
four scores are reported, the average of reported scores is used. 
The multinomial logit model of school choice examines the school which each 
individual attended (ORDER).  This is a categorical variable representing public, 
catholic, other religious private, non-religious private and non-accredited schools.  This 
multinomial logit model is used to explain why parents send their children to a non-
public school.  The income model is measured by the individuals? income in year 1999, 
 51
which is at age 25 for most students (AINCOME).  This variable ranges from a zero 
income to a 500,000-dollar income.  This model estimates the impact of education on 
annual income.       
Following the literature, the variables include student?s gender (MALE, 
FEMALE), race/ethnicity (ASIANPI, AMERIN, HISPAN, BLACK, WHITE), parental 
highest education levels (FATHSCH and MOMSCH), total family income (FAMINC), 
student-teacher ratio (STRATIO), and percent of students on free or reduced lunch 
(PCTLUN), total school enrollment (TENROLL) and student?s own educational 
achievement (RHDEG).  
Summary Statistics 
 Out of a total of 12,144 observations in the dataset, 5,600 observations lack data 
on student performance measurements at the high school level (POSIM); approximately 
2,600 or more observations have missing data on one or more of the exogenous variables, 
namely, PCTLUN and SALHIGH; yet another 900 observations miss information on the 
instruments STRATIO and TENROLL.  As discussed in Xuejuan (2004), observations 
with missing dependent variables do not contribute to estimation efficiency; I drop the 
5,600 observations from the sample for this specific model.  The actual sample used for 
estimation has 6,544 observations, or 53.9% of the original dataset. 
 The school choice and income estimations are a different story.  Out of 12,144 
observations, only 3,815 lack data on income at age 25 (AINCOME), and only 1,738 
observations lack data on the school choice at the high school level (ORDER).  The 
variables, UNDGPA and RACECUR, from the income estimation together have about 
3,100 observations that have missing information.  The variable STRATIO has 
 52
approximately 900 observations missing data, affecting the school choice model.  The 
actual sample used for the income model is about 69.1% of the original dataset, and about 
68.5% for the school choice measurements.  The summary statistics are reported in Table 
2. 
 As can be seen, 47.6% of the students are male; 7.0% Asian/Pacific islander, 
1.3% American Indian, 13.4% Hispanic, 9.65% Black; and 68.4% White.  About 68.4% 
of the students? fathers have more than a high school degree and about 72% of the 
mothers also have more than a high school degree.  About 30% of the students did not 
have a college degree by the age of 25.  On family income about 38.3% have between 
10,000 and 34,999 dollars and about 33.2% between 35,000 and 99,999 dollars.  
Student?s annual income averaged $27,400 in 1999.  Attendance was 84.3% in public 
schools, 5.86% catholic, 3.5% private, and 2.1% other religious private schools.  About 
27% of the high schools are in urban areas, 39% are suburban and 30% are in rural areas; 
and average total school enrollment is approximately 656 students. In the eighth grade, 
the average student scored at the 51.4 percentile in his standardized test and the twelfth 
grade the average student scored 51.2; and on average there is a -.54 of decline in test 
scores between the years 1988 and 1992. Seven years after high school, about 31.7% 
students had completed a Bachelor?s degree and about 11.2% had obtained some type of 
graduate degree.
 
 53
Table 2 Summary Statistics 
Variable          Mean          Std. Dev.       Min        Max 
Male            .4761            .4995         0           1 
Female           .5238            .4995         0           1 
Asianpi          .0703            .2557         0           1 
Amerin          .0131            .1140         0           1 
Hispan           .1335            .3401         0           1 
White           .6842            .4648         0           1 
Black            .0965            .2954         0           1 
Public            .8429            .3638         0           1 
Cath            .0586            .2349         0           1 
Othrel           .0211            .1439         0           1 
Priv             .0351            .1839         0           1 
Urban           .2733            .4456         0           1 
Subur            .3862            .4869         0           1 
Rural            .3023            .4593         0           1 
Northe           .1838            .3873         0           1 
Midw            .2590            .4381         0           1 
South            .3259            .4687         0           1 
West            .1934            .3950         0           1 
Pctlun               20.4113              20.8984                0          100 
Salhigh            42051.61              9264.22            12000       78000  
Stratio               17.7271                 5.1269        5          50 
Test 92               5120.76             894.0281           2935       6977.7 
Test 88               5140.58             894.8566           3232       7545.5 
Nodeg (res. Ed)     .3023            .4593         0           1 
Unddeg          .3177            .4656         0           1 
Grddeg          .1116            .3149         0           1 
Mths (father)       .6839            .4649         0           1 
Lths (father)       .1521            .3592         0           1 
Mths0 (mother)     .7204            .4488         0           1 
Mlths (mother)     .1479            .3550         0           1 
Tenroll               656.501              374.0995                6          3940 
Ten (faminc)       .0917            .2885         0           1 
Thirtf            .3833            .4862         0           1 
Hund            .3318            .4709         0           1 
Mhund           .0453            .2079         0           1 
Posim              -54.6412             464.9783               -3001        2875.5 
Pnorth           .0236            .1519         0           1 
Psouth           .0115            .1064         0           1 
Undgpa                2.6406            .8502         0           4 
Aincome         27.38            22.44         0          500 
Raceor               64.1951                 31.8785               0          100 
Racecur              75.0222                        30.6496               0          100 
 
 
 54
 
CHAPTER IV 
 
MODEL RESULTS 
 
Variables 
 The student performance estimation contains variables measuring the type of high 
school, namely whether the high school is public, private, catholic, or other religious 
private, the community in which the high school is located (URBAN, SUBUR, and 
RURAL), the school average student/teacher ratio (STRATO), the average total student 
enrollment (TENROLL), the highest salary paid to teachers (SALHIGH), to emphasize 
the importance of the teacher?s education, and the percentage of students in the high 
school that are on free or price reduced school lunch (PCTLUN).   
 The quality of high school is hypothesized to influence a parents? choice.  Parents 
may choose a high school based on their housing location or they clearly choose a private 
school.  Endogeneity of the variables may be a problem.  The first variable to study is the 
choice between a public and a private school (ORDER).  There are different types of 
private schools, but when parents are choosing a school quality for their child, the 
academic performance is not the only factor considered in the decision.  Other factors 
include religious reasons, location of school, extracurricular activities, percentage of 
minority students, etc.  Class size is measured as the average number of students per 
teacher in the school (STRATIO).  Endogeneity occurs with this variable when a 
particular student is placed into a class for a certain reason.  For instance, a child may be 
placed into a smaller class based on his or her lack of ability.  The student/ teacher ratio is
 
 55
 
 not correlated with the school choice variable making endogeneity less likely a problem.  
Lastly, the high school community (URBAN, SUBUR, and RURAL) can be a factor in 
the endogeneity problem.  It is expected that an average suburban high school provides a 
better school quality than an average rural or urban high school.  But for the parents it is 
more than likely better to choose a school within a certain location than to choose a 
school from all three categories (URBAN, SURBUR, RURAL), which reduces the 
endogeneity problem.  Therefore, endogeneity does not appear to be a problem. The 
correlation matrix does not reveal any variables with a correlation above 0.5000. 
Tests for Models 
 The student performance model and the income model were tested for 
heteroskedasticity and specification error.  Autocorrelation was not tested in these 
models, because time series analysis was not conducted.  The Breusch-Pagan/Cook ? 
Weisberg test was used to test for heteroskedsticity.  The null hypothesis states that the 
regression analysis portrays homoskedasticity and the alternative is that it does have 
heteroskedasticity.  The student performance model Chi-square was equal to 12.55, 
rejecting the null hypothesis suggests in that this model does reveal a heteroskedasticity 
problem.  The income model heteroskedasticity test shows that the Chi-square was equal 
to 954.35.  The probability of the function being greater than the chi-square is nearly 
zero.  This implies that the income model does have a heteroskedasticity problem.  
Therefore, we reject the null hypothesis.  To correct for heteroskedasticty, weights from 
the sampling data were used.   
  
 
 56
 
 The Ramsey RESET test was used to test for specification error.  The null 
hypothesis states that the model does not have specification error and the alternative is 
that is does have specification error.  This test was used for both the student performance 
and income model.  The income model specification error test shows that F test statistic 
of 8.60, compared to a critical value of 1.62 at the 5% level.  Therefore, we reject the null 
hypothesis.  The student performance model specification error test demonstrates that the 
F test statistic of 1.30, so we fail to reject the null hypothesis.    
 The student performance model contains the state dummy variables to capture the 
school quality effects from each individual state.  I performed an F-test to determine 
whether or not it is important to include the state dummy variables.  The null hypothesis 
states that the state characteristics are jointly equal to zero and the alternative states that 
the individual state dummy variables are not jointly equal to zero.  The F-test statistic is 
calculated as follows:  
 F-test = R? (full model) - R? (reduced model)/ (1-R?) * (n-k)/J  
 n= number of observations, k= degrees of freedom and j= number of restrictions 
 F-test = (0.0408 ? 0.0158)/ (1-0.0408) * (6508-69)/49 
 F-test = 0.0260 * 131.408 = 3.42 
 Critical F (49, 6000) = 1.50 
F > critical F, therefore we reject the null hypothesis.  
The purpose of this f-test is to demonstrate the importance of the individual state.  The 
characteristic of each individual state is an important factor in student performance.   
 
 
 
 57
 
School Choice Estimation 
  A multinomial logit model was used for the school choice estimation.  A linear 
model was used for the other two models.  The multinomial logit model was applied to 
examine the reason why students attend a private school.  This model emphasizes the 
factors that parents consider when choosing a private school.   
 Table 3 reports the school choice estimation results.  The dependent variable 
(ORDER) is a categorized instrument sorted by types of high schools.  The base school 
type is represented by public schools (number 0).  The existing literature was used as a 
guide in choosing the exogenous variables for this estimation (Hanushek 2004).  These 
variables include:  Sex of student (FEMALE), race/ethnicity (HISPAN, BLACK, 
ASIANPI, and AMERIN), the region of the United States (NORTHE, MIDW, SOUTH) 
and the community of the school (URBAN and SUBUR), the composite of the 
student/teacher ratio (STRATIO), mother?s and father?s educational attainment (MTHS, 
MLTHS) and family income (TEN, THIRTF, and HUND).  Table 1 shows definition of 
variables. 
 According to Table 3, the base categories for dichotomous variables are the white, 
male, rural, the western states, the student?s father and mother with less than a high 
school education, and annual family income of more than 100,000 dollars.  Family 
background such as father?s educational level (more than high school), is a very 
important factor for a child attending a catholic school.  Only the father?s educational 
levels shows a significant effect on catholic school enrollment.  Family income of below 
100,000 dollars shows a negative impact on catholic school attendance.  This portrays 
that families with a high level of income will be more likely to send their child to a 
 
 58
 
private school.  This estimation demonstrates that a black female living in a rural area has 
a smaller possibility to attend a catholic school than a white male living in an urban or 
suburban area.  The community and location of the school play important roles when 
deciding on a school for a child.  All regions of the United States show a significant 
effect on catholic school enrollment when compared to the omitted region, west.  The 
variable student/teacher ratio (STRATIO) demonstrates the school quality.  This variable 
reveals a positive significant effect.  This suggests the higher the number of students per 
teacher for the school.  Therefore, the higher the possibility a child is to attend a catholic 
school. 
Table 1 Definition of the Variables 
MALE  Gender of student 1 male, 0 female 
FEMALE  Gender of student 1female, 0 male 
ASIANPI  Whether student is Asian/Pacific Islander, 1yes, 0 no 
AMERIN  Whether student is American Indian, 1yes, 0 no 
HISPAN  Whether student is Hispanic, 1 yes, 0 no 
BLACK  Whether student is Black 1 yes, 0 no 
WHITE  Whether student is White 1 yes, 0 no 
URBAN  Whether student?s high school is central city, 1yes, 0 no 
SUBURBAN  Whether student?s high school is suburban area, 1yes, 0 no 
RURAL  Whether student?s high school is rural area, 1 yes, 0 no 
PUBLIC (ORDER) Whether student?s high school is public, 1yes, 0 no 
CATH  Whether student?s high school is catholic, 1 yes, 0 no 
OTHREL  Whether student?s high school is other religious private, 1 yes,  
  0 no 
PRIV  Whether students high school is a private (non-religious), 1yes,  
  0 no 
PCTLUN  Percentage of students on free or reduced price lunch in high 
  School 
TENROLL  Total number of students in the high school 
SALHIGH  The highest salary paid to teacher?s in the high school 
TEST 88  Standardized test scores in the eighth grade, average among  
reading, mathematics, science, and   
history/citizenship/geography 
TEST 92  Standardized test scores at the twelfth grade, average among  
reading, math, science, and history/citizenship/geography 
  
 
 
 59
 
Table 1 Continue 
POSIM  12
th
 ? 9
th
 grade average standardized test scores which captures 
the positive improvement between two times 
STRATIO  Composite student/teacher ratio in high school 
UNDGPA The undergraduate GPA of each student 
UNDDEG  Whether the student graduated from college by the year 2000,  
  1 yes, 0 no 
GRDDEG Whether the student completed a master?s or higher degree 
 by the year 2000, 1 yes 0 no 
NODEG Whether the student graduated from high school by the year  
 2000, 1 yes, 0 no 
ANICOME The annual income (thousand dollars) of student at age 25, 1999 
RACECUR Students estimate of what percent of people in their current  
 Neighborhoods of the same race as they are 
NORTHE Whether student?s high school is located in the Northeast region  
 Of the United States, 1 yes, 0 no 
MIDW Whether student?s high school is located in the Midwest region  
 of the United States, 1 yes, 0 no 
SOUTH    Whether student?s high school is located in the Southern region 
of the United States, 1 yes, and 0 no 
MTHS (FATHSCH) Whether the student?s father continued beyond high school, 1 
yes, 0 no 
LTHS            Whether the student?s father graduated from high school, 
                1 yes, 0 no 
MTHS0 (MOMSCH) Whether the student?s mother continued education beyond high 
school, 1 yes, 0 no 
MLTHS           Whether the student?s mother graduated from high school,  
                1 yes, 0 no 
TEN (FAMINC)     The total family income is less than 10,000 dollars, year 1988, 
                1 yes, 0 no 
THIRTF           The total family income is between 10,000 and 34,999 dollars,  
                year 1988, 1 yes, 0 no 
HUND            The total family income is between 35,000 and 100,000 dollars, 
                year 1988, 1 yes, 0 no 
PSOUTH          Whether the students who attended private school are from  
                Alabama, Georgia, Mississippi, Louisiana, and Tennessee, 1 yes, 
                0 no 
PNORTH          Whether the students who attended private schools were from  
                states other than southern states, 1 yes, 0 no 
DVSTATES        The states in which the students were located in year 1992,   
                1 yes, 0 no (i.e. dval, dvar, dvaz, dvga, dvpa, dvms, dvla, dvtn) 
 
 60
 
Table 3 School Choice Model Estimation Results   
Dependent Variable                Whole Sample 
ORDER           Coef.            T-stat             P-value       
Catholic School 
Gender 
Female student        -.3914                -2.50              0.012 
School Quality 
Student teacher ratio      .1805               10.94                 0.000 
Race 
Asian Pacific Islander       .1205                0.39             0.697 
American Indian          .2675                  0.33             0.741 
Hispanic               -.1624               -0.70             0.487 
Black              -.8006                -2.47             0.014 
Community 
Urban area                 21.0700                      44.84             0.000 
Suburban area                18.8298                        35.83                0.000 
Region of the US 
Northeast             1.9428                 8.14              0.000 
Midwest           1.6978                   7.83              0.000 
South            1.2498                  5.15             0.000 
Parents Education 
Father (more than HS)    .5039               2.35              0.019 
Mother (more than HS)     .0443                      0.17              0.861 
Family Income 
10,000 or less               -1.8705                      -5.18              0.000 
10,000 - 34,999           -.9357             -5.18              0.000 
35,000- 99,999            -.2338                 -1.16              0.244 
Constant                   -27.1465              -                   -        
Private Other Religious 
Female           .4313            1.57               0.117 
School Quality 
Student Teacher ratio      .0776           2.71               0.007 
Race 
Asian Pacific Islander     -.7250            -1.76              0.078   
American Indian             2.5210           2.89               0.004  
Hispanic           -1.6812            -4.12               0.000 
Black                -1.9633                   -2.11               0.035  
Community 
Urban area                 1.9202           4.20               0.000 
Suburban area         .8999            2.07               0.039 
Region of US 
Northeast           .7729            1.64               0.101 
Midwest                 -.2386                   -0.68               0.499 
South           .8513           2.410               0.016       
 
 
 
 61
 
Table 3 Continue 
Dependent Variable                Whole Sample 
ORDER           Coef.            T-stat             P-value       
Parent Education 
Father (more than HS)    .5018            1.18              0.236 
Mother (more than HS)   .9721                      2.34              0.019 
Family Income 
9,999 or less               -33.8295                  -86.95             0.000 
10,000 ? 34,999             -1.0329               -2.60              0.009 
35,000 ? 99,999                -.3992             -1.25              0.212 
Constant                 -7.5495                    -6.93              0.000      
Private Non-religious 
Female           -.3206            -1.20              0.229 
School quality 
Student teacher ratio      -.1801              -2.00             0.046 
Race 
Asian Pacific Islander         .0535                0.18             0.856 
American Indian          -38.9859           -128.70              0.000 
Hispanic             -2.0466              -4.06             0.000 
Black              -1.6581                     -2.36             0.018 
Community 
Urban area            1.7867              6.16                 0.000 
Suburban area            -.1738             -0.78                0.436 
Region of US  
Northeast            .4942                 0.76              0.447 
Midwest                   -.3602            -0.81              0.420 
South          1.3079                2.61              0.009 
Parents Education 
Father (more than HS)   1.4832              5.13             0.000  
Mother (more than HS)     .9082                      3.49             0.000 
Family Income 
9,999 or less              -2.6247                      -3.39              0.000 
10,000 ? 34,999           -2.2547              -8.48              0.000 
35,000 ? 99,999            -.9899              - 4.62              0.000 
Constant               -3.2052               -1.72             0.085        
Non-Accredited  
Female          -.0090                      -0.01             0.990 
School Quality 
Student teacher ratio      .1278               1.49             0.137 
Race 
Asian Pacific Islander     -.4928             -0.64             0.523 
American Indian           -38.8133             -82.20             0.000 
Hispanic           .5240               0.55             0.579        
 
 
 
 
 62
 
Table 3 Continue 
Dependent Variable                Whole Sample 
ORDER           Coef.            T-stat             P-value      
Black         -2.2570           -2.09              0.037 
Community 
Urban area           1.1264              1.65             0.098 
Suburban area           .5145                 0.66              0.509 
Region of US 
Northeast            2.4198               1.75             0.079 
Midwest              .6122               0.80             0.427 
South            1.6029            1.80              0.072 
Parents education 
Father (more than HS)     .9300             1.54             0.123 
Mother (less than HS)       .8399                   1.25              0.213 
Family Income 
9,999 or less              -.7276                   -0.61             0.545 
10,000 ? 34,999         .7288             0.77             0.441 
35,000 ? 99,999         .1540             0.30             0.762 
Constant              - 11.2488                  -3.37             0.001         
Outcome order =0 (public schools) is the comparison group 
Number of Obs. = 10406 
R-Squared= 0.2838 
 
 Next, is the evaluation of attendance of other religious private schools.  In this 
estimation Hispanic (HISPAN), Asian Pacific Islander (ASIANPI), and Black (BLACK) 
students show a negative effect and American Indian (AMERIN) students demonstrate a 
positive effect on religious private school attendance.  A female student living in a urban 
(inner city) or suburban area has a higher probability of attending a religious private 
school, compared to a male living in a rural area.  The southern region of the United 
States (SOUTH) is the only region variable that demonstrates a positive and significant 
effect.  Family income between 34,500 and 99,999 dollars demonstrate a higher 
possibility of sending their child to a religious private school.  A family with an annual 
income of 100,000 dollars or more shows no significance.  A mother?s educational 
attainment of more than high school reveals a positive and significant effect.  The 
education of the parent is an important factor in choosing a school type.  The school 
 
 63
 
quality variable, student/teacher ratio, illustrates a positive and significant effect on 
student attending a religious private school.  
 The third school type is the private school without any religious affiliation.  This 
demonstrates American Indian (AMERIN), Hispanics (HISPAN), and African Americans 
students (BLACK), are less likely to attend a non-religious private school, compared to 
white students.  A student living in an inner city (urban) area and located in the southern 
part of the United States has a higher probability of attending a non-religious private 
school, compared to the rural area and other regions of the United States.  Family income 
has the same outcome as demonstrated previously.  However, the student/teacher ratio 
variable shows a negative significant effect.  The father?s and mother?s education level 
has a positive effect.  
 The last school analysis is the non-accredited schools.  This analysis reveals that 
American Indian, Asian Pacific Islander, and African American students all demonstrate 
a significant negative effect on non-accredited school attendance, compared to white 
students.  A student living in a suburban area and in the northeast or southern region of 
the United States has a lower probability of attending a non-accredited school.  The 
school quality variable, student/teacher ratio reveals no significance on non-accredited 
school attendance.  Neither parent?s educational level portrays any significant effect.  
However, the family income variables demonstrate positive effects, except for the less 
than 9,999 dollar range, however they are not significant.  
 
 64
 
Student Performance Model Results 
 Table 4 and 4a report the regression analysis results.  In Table 4 the dependent 
variable (POSIM) is the average standardized test score in all four subject areas for 1992 
(twlelfth grade) minus test scores from the year 1988 (eighth grade).  This variable 
demonstrates the change in standardized test scores between two time periods.  Following 
the existing literature, the exogenous variables include: sex of student (MALE), 
race/ethnicity (ASIANPI, AMERIN, HISPAN, and BLACK), school types (CATH, 
OTHREL, PRIV, and NACRED), the community of the school (SUBUR and RURAL), 
the percentage of students on free or price reduced school lunch (PCTLUN), highest 
salary paid to teachers (SALHIGH), school wide student/teacher ratio (STRATIO), 
mother and father?s educational attainment (MTHS and MTHS0), total student 
enrollment (TENROLL), family income (TEN, THIRTF, and HUND), the state that the 
school is located (DVSTATES, DVAL, DVMS, etc. ) and whether the  private school is 
located in the southern and northern states (PSOUTH and PNORTH).  In table 4a, instead 
of including the private schools located in the southern and northern states I included the 
private school type.  This will illustrate the important factors that effect school quality in 
the private schools in the southern states and explain if parents are sending their children 
to private schools for other than academic reasons. 
 According to Table 4, male students have a significant positive increase of .51 
percentile in performance, when compared to females.  The only significant race variable 
is Asian Pacific Islander and they show a positive increase of .90 percentile.  Catholic 
schools are the only significant variable of the school type.  Catholic schools demonstrate 
a 1.16% increase in performance.  The total enrollment variable shows a .0006% positive 
 
 65
 
impact on the dependent variable.  This explains that the larger the school the higher the 
improvement on test scores, which illustrates those bigger schools provide more 
resources for the students.  The schools in Louisiana have a 1.85% positive improvement 
on test scores.  The schools in Alabama, California, Connecticut, Kentucky, Montana, 
Oklahoma, Pennsylvania, Tennessee, Texas, and Utah all have a significantly negative 
effect on test scores.  Private schools in the southern and northern states show a positive 
impact on test scores, but neither is significant. 
 Table 4a shows the student performance estimation with the private school 
variable (PRIV), instead of the variables that include the private schools located in the 
southern and northern states (PSOUTH and PNORTH).  In this analysis the two 
coefficients PSOUTH and PNORTH were tested to see if they were equal.   
 H
o
:  ?
19
 = ?
20 
 
H
a
:  ?
19
 ? ?
20 
The F-test is as follows: 
 F-Test statistic = 0.30 
 F (1, 6437) = 3.84 
 Probability of F > 0.30 = 0.5853 
 F>F
c 
 Therefore, we fail to reject the null hypothesis that the coefficients of PSOUTH 
and PNORTH are not equal to one another.  Both models show that on average private 
schools, except for catholic, do not provide any better or any worse of an education than 
public schools.  
 
 66
 
Table 4 Student Performance Model Estimation Results 
(Including variables psouth and pnorth) 
Dependent variable    (percentile)           Whole sample 
POSIM            Coeff.                   T-Stat                  P-Value       
Private schoolsIn the South 
Psouth               .7016              0.93              0.350 
Pnorth                .0526                 0.05              0.957 
Male Student          .5123                       2.47                     0.014 
Race 
Asian Pacific Islander     .9009                      2.83              0.005 
American Indian             -.5118              -0.67              0.505 
Hispanic                       .6207               1.34              0.181  
Black                       -.7481              -1.10              0.270  
School Type 
Catholic                   1.1662               2.56              0.011 
Other Religious         .4173              0.84              0.402  
Community 
Subur ban area          .1296               0.36              0.721 
Rural area             -.0457                    -0.12              0.907 
School Quality 
Percent Lunch              .0063                           1.15                     0.249 
Highest Salary Paid        .0000               1.51              0.131 
Student teacher ratio      .0237                              1.14              0.256 
Tenroll                   .0006              1.70              0.090 
Parent Education 
Father (more than HS)    .3126                    1.09                  0.276  
Mother (less than HS)      .1818               0.49                 0.627 
Family Income 
10,000-34,999         .1318               0.37                0.714 
35,000 ? 99,999        .3121                1.02                0.307 
9,999 or less          .3726               0.84                0.399 
State Variables 
West 
WA                       -.2937               -0.43              0.669 
OR                      .2338                0.25              0.801 
CA                              -1.4318                             -2.46              0.014 
MT                    1.4373                  0.83              0.409 
ID                          .2591                                0.30              0.761 
NV                  -1.7935                -0.82              0.411 
UT                -2.0831                -2.47              0.014 
AZ                                 -.7588                              -0.68                             0.495 
WY                -1.3811                -1.11              0.265 
CO                              -81.2477                              -1.01              0.314 
NM                 -.3273               -0.40              0.687 
HI                      .4529                    0.50                0.618 
 
 
 67
 
Table 4 Continue 
Dependent variable                 Whole sample 
POSIM            Coeff.                   T-Stat                  P-Value    
Mid-West 
ND                  .3157                   0.36            0.717 
SD                   -.8894                 -1.00            0.319 
NE                  -.4593                  -0.58            0.559 
KS                      -.0429                                 -0.05            0.958 
OK               -2.3653                 -1.98            0.048 
TX              -1.8183                                 -1.74              0.081  
MI                           -.563105                   -0.83            0.406 
IA                -2.003924                                 -2.45             0.014 
MO             -1.322634                   -2.04            0.041 
IL                 -1.553015                                 -2.44              0.015 
WI               -1.035938                  -1.59            0.111 
MN                    .010651                     0.01             0.988 
Northeast 
KY                - 2.2922                               -3.23                0.001 
VA                 -.7576                 -1.05             0.293 
OH                 -.9825                -1.62             0.106 
PA                -1.4403                 -2.32             0.020 
NY                 -.3779                -0.58             0.562 
WV              -1.2888                 -1.49             0.136 
MD                             .1229                                 0.17             0.866 
DE                           - 1.3720                                 -1.18             0.236 
DC                              .5288                                0.55             0.584 
NJ                -1.1048               -1.38             0.169 
CT                             -1.7169                              -1.96             0.050 
RI                 -1.1065               -1.07             0.284 
MA                             -.4193                               -0.56             0.576 
VT                 2.1605                1.36             0.173 
NH                -.6639                -0.74             0.459 
ME                            -.1215               -0.06             0.950 
IN                 -1.8818                              -2.75             0.006 
South 
AL                      -1.2004                -1.78             0.075 
AR                    .0765                    0.10             0.920 
FL                              -.7070                               -1.08             0.280 
GA                     .2566                     0.35                         0.729  
LA                             1.8591                                 1.84                 0.066 
MS                   -.5535                  -0.71            0.480 
NC                  -.2761                   -0.24             0.808 
SC                 -.0745                    -0.10             0.923 
TN               -1.4950                   -2.15             0.031 
Constant           -2.4165                -2.01             0.044     
R-squared-   0.0408           Number of Obs. - 6508 
 
 68
 
Table 4a Student Performance Model estimation Results (with variable Priv) 
Dependent Variable                   Whole Sample 
POSIM             Coef.            T-stat            P-value        
School Type 
Private               .2320                 0.31              0.756 
Catholic           1.1675                 2.56              0.011 
Other religious         .4185                 0.84              0.401 
Gender 
Male                .5121                 2.47              0.014 
Race 
Asian Pacific Islander    .9005                2.83              0.005 
American Indian          -.5081               - 0.66                0.508 
Hispanic              .6234                  1.35              0.179 
Black                        -.7533               - 1.11              0.265 
Community 
Suburban area           .1289                 0.36              0.722 
Rural area             -.0444              - 0.11              0.909 
School Quality 
Total enrollment         .0006               1.70              0.089 
Percent Lunch           .0062               1.14              0.256 
Highest salary paid         .0000                 1.51              0 .131 
Student/teacher ratio      .0236               1.13              0.257 
Parent Education 
Father (more than HS)  .3115               1.09              0.277 
Mother (less than HS)    .1815                0.49              0.627 
Family Income 
9,999 or less                   .3770                               0.85              0.393 
10,000 ? 34,999            .1352              0.37              0.708 
35,000 ? 99,999                .3149             1.03              0.302 
State Variables 
West 
WA                    -.2790                             -0.41             0.684 
OR                     .2480                0.27             0.789 
CA                             -1.4190                 - 2.44             0.015 
MT                         1.4516                0.84             0.403 
ID                                .2725                  0.32            0.749 
NV                         -1.7797              -0.82             0.414 
UT                         -2.0681                              -2.45            0.014 
AZ                             -.7448                 0.67            0.503 
WY                         -1.3786                              -1.11            0.267 
CO                            -.80102                 -1.00             0.320 
NM                          -.3136              -0.39             0.700 
HI                              .4660                    0.51            0.607 
Mid- West 
ND                    .3286                 0.38              0.705 
SD                         -.8737                                -0.98               0.326       
 
 69
 
Table 4a Continue 
Dependent Variable                   Whole Sample 
POSIM             Coef.            T-stat            P-value       
Mid-West 
NE                         -.4448                 -0.57           0.570 
KS                -.0295                  -0.04             0.971 
OK                        -2.3506                  -1.97            0.049 
TX                             -1.8183                                 -1.74            0.081  
MI                      -.5479                    -0.81            0.410 
IA                            -1.9898                 -2.44            0.015 
MO                     -1.3088                -2.03            0.042 
IL                            -1.5400                  -2.42            0.015 
WI                       -1.0215                             -1.58            0.115 
Northeast 
KY                     -2.2767                 -3.22             0.001 
VA                        -.7435                             -1.04              0 .300 
IN                    -1.8674                -2.73             0.006 
MN                 .0251                   0.03             0.973 
OH                         -.9692             -1.60             0.109 
PA                       -1.4266                             -2.30             0.021 
NY                         -.3645               -0.56             0.575 
WV                       -1.2748                             -1.48             0.140 
MD                 .1383                   0.19            0.849 
DE                           -1.3561                 -1.17             0.241 
DC                             .5472                   0.57             0.568 
NJ                        -1.0907                 -1.36             0.174 
CT                            -1.7060                 -1.95             0.052 
RI                         -1.0919                              -1.06            0.290 
MA                       -.4152                 0.55            0.579 
VT                        2.1747                               1.37            0.170 
NH                         -.6882                -0.77            0.440 
ME                       -.1080                 -0.06            0.955 
South 
AL                             -1.1941                      -1.76            0.078 
AR                              .0905                 0.12            0.905 
FL                            -.6950                -1.06            0.287 
GA                              .3289                    0.45            0.650 
LA                     1.8786                    1.87            0.062 
MS                  -.4244                -0.56             0.574 
NC                    -.2620               -0.23            0.818 
SC                    -.0578                              -0.08            0.940 
TN                       -1.4664                              -2.13            0.033 
Constant                  -2.4296                 -2.03             0.043         
Number of Obs = 6508 
R-squared = 0.0407 
 
 70
 
Income Model Estimation 
 Table 5 and 5a report the regression analysis results.  In table 5 the dependent 
variable (AINCOME) is the annual income in thousands of the student at about age 25 
for the year 1999.  The exogenous variables for this model are as follows:  gender of 
student (MALE), school types (CATH and OTHREL), the race/ethnicity of the student 
(ASIANPI, AMERIN, HISPAN, and BLACK), percentage of the population in the 
students current neighborhood with the same race as they (RACECUR), region of the 
United States that the school is located (NORTHE, MIDW and SOUTH), highest degree 
earned by the student, either undergraduate or graduate degree (UNDDEG and 
GRDDEG), student performance measured by standardized test scores for year 2000, 
twelfth grade (TEST 92), and dummy variables for private schools located in the southern 
and northern states (PSOUTH and PNORTH).  In table 5a, instead of the TEST 92 
variable the undergraduate GPA of the student is included (UNDGPA).  
 According to table 5, if a student attends a catholic school it will raise income by 
3,200 dollars per year, but if a student attends a private school located in a northern state 
it will decrease the income by 700 dollars, but this is not a statistically significant effect.  
This estimation establishes that males have approximately a 7,800 dollars higher income 
than females.  If a student is an Asian Pacific Islander or Hispanic it decreases income by 
about 2,500 dollars, but if the student is Black then it decreases the income by 
approximately 3,000 dollars.  A student who chooses to start a career in the south 
decreases income by approximately 2,100 dollars.  A student that has obtained an 
undergraduate degree or graduate degree increases income by about 3,600 dollars. This 
implies that additional education can be an important factor in raising income.  A student 
 
 71
 
who did well on the standardized test scores in the twelfth grade raises income by 25 
dollars a year.  On the other hand, this estimation demonstrates that attending a private 
school does not affect income.  The PSOUTH variable was not significant, but has a 
coefficient of 1,200. 
 Table 5a demonstrates the income model including the variable UNDGPA and 
excluding TEST 92.  This table evaluates the importance of post-secondary education on 
annual income.  A student who graduates from college can increase income by about 
4,800 dollars, but if you graduate with a master?s or higher you can increase income by 
5,700 dollars.  The undergraduate GPA variable did not show any significance on annual 
income.  Theses tables illustrate that attending a private school can not increase or 
decrease your income anymore than attending a public school.  However, the catholic 
school has been the only school that has show significant and positive effects on income.   
Table 5 Income model Estimation results (with variables Test 92, Psouth and Pnorth) 
Dependent Variable  Whole Sample 
AINCOME           Coef.            T-stat            P-value       
Private schools in South 
Pnorth                -.7147               -0.11            0.911 
Psouth                 1.2658               0.63            0.531 
Student Performance 
Test 92                  .0025                 5.66                  0.000 
School type 
Catholic                     3.2852                 2.02            0.043 
Other religious          -.9611              -0.51             0.609 
Gender 
Male student           7.8985              12.37            0.000 
Race 
Asian Pacific Islander     -2.7717              -1.91            0.056 
American Indian       -3.6682              -1.48            0.139 
Hispanic             -2.1160             -2.11            0.035   
Black               -3.3046             -3.47            0.001 
Regions of the US 
Northeast            -1.7482             -0.61            0.543 
Midwest             -1.1608             -1.01            0.313 
South               -2.1333             -1.90            0.058       
 
 72
 
Table 5 Continue 
Dependent Variable                       Whole Sample 
AINCOME           Coef.              T-stat           P-value       
Personal Attributes 
Obtained Grad Degree       3.6501              3.69             0.000   
Undergraduate Degree     3.7594               5.01                   0.000 
Current Race in area        -.0190             -1.51             0.131 
Constant               13.0088               4.65            0.000        
Number of Obs. = 8321            
R-squared = 0.0721 
F(17,  8303) =  28.47                                               
 
Table 5a Income Model Estimation Results 
(with variables Psouth, Pnorth, and Undgpa) 
Dependent Variable                Whole Sample 
AINCOME           Coef.          T-stat            P-value          
Student Performance 
Undergraduate GPA       -.3679          -0.87            0.382 
Private Schools in South 
Pnorth               2.1346            1.12            0.262  
Psouth               1.2620           0.22            0.828 
Gender 
Male student           7.3927          11.87            0.000 
School type 
Catholic              3.8587            2.61            0.009 
Other religious        -.8213           -0.46            0.645   
Race 
Asian Pacific Islander       -1.1622         -0.64            0.520 
American Indian             -10.3411         -3.89              0.000 
Hispanic              -2.5589         -2.36            0.018 
Black                -3.8432         -4.23            0.000 
Regions of the US 
Northeast                -.4315         -0.39            0.700 
Midwest                 -.8605         -0.84            0.401 
South                -1.8476         -1.80            0.072 
Personal Attributes 
Current Race in area          -.0104         -0.84            0.400 
Obtained Grad. Degree      5.7158          6.19            0.000 
Undergraduate Degree        4.8021            6.33            0.000 
Constant                  29.0076         16.17           0.000            
Number of Obs. = 7777               F (17, 7759) = 15.14 
R-squared = 0.0491 
 
 73
 CHAPTER VI 
SUMMARY AND CONCLUSION 
 
 This paper proposes three individual models to identify school choice, student 
performance, and annual income of a cohort of students.  The school choice model tested 
factors thought to influence a parent?s choice of a school for their child.  The student 
performance model examines the impact of variables on standardized test scores.  The 
income model equation tests the effects of the student?s education (UNDGPA) and 
student performance (TEST 92) to determine its influences on annual income.   
 In the school choice model, the p-values of the relative variables generally 
resulted with the expected signs and magnitudes.  The mother?s and father?s education 
plays an important role in the child attending a non-public school, as expected.  The 
gender and race/ethnicity of the student explained why students attended certain schools.  
For example, a Hispanic male student that lives in an urban or suburban area located in 
the Northeast, Midwest or South of the United States will be more likely to attend a 
catholic school.  The location of the school plays an important role which follows the 
existing literature.   
 In the student performance model there were few significant explanatory 
variables, possibly due to missing data information.  The significant variables in this 
model did represent the gender and race/ethnicity of the student, school type, and the total 
enrollment of the school. In the results an Asian Pacific Islander male student attending a 
large school system is expected to have a greater improvement on standardized test 
 
 74
scores between two time periods.  These results explain that larger schools provide more 
resources, therefore students in theses schools have a higher improvement on 
standardized test scores.  Private school attendance was not significant in this estimation. 
 The income model shows many significant variables.  This model demonstrates 
that white males make more money than females, and that other races will have a lower 
income.  The location of the student does affect income; if a student lives in the south it 
will decrease annual income, possibly because the cost of living is lower than in the 
northern part of the United States.  A student?s educational attainment plays a critical role 
in annual income.  Achieving an undergraduate or graduate degree can increase annual 
income.  The only school type that was significant was the catholic school.  The private 
schools in the south or north do not show any significant effect on annual income.  
 These results explained many factors that affect income, student performance and 
school choice attendance.  But primarily school choice is a parental choice and they 
choose schools based on many factors besides academic performance.  However, these 
parents who are choosing private schools are not necessarily providing their child with a 
better education, because on average a private school does not provide a better education 
than private schools.  The improvement of education is the main goal.  The improvement 
of education cannot only better Alabama, but also better the United States? financial 
strength, health and technology.  
 
 75
BIBLIOGRAPHY 
Ashley, C. F.  ?Integration.?  The Columbia encyclopedia.  Sixth edition.  New  
  
 New York:  Columbia University Press, 2001-05. 
 
Averett, L.S. and Burton, L.M. ?College Attendance and the College Wage Premium:  
  
 Differences by Gender.? Economics of Education Review. 15 (1996):37-49. 
 
Belfield, R.C. ?The Effects of Competition on Educational Outcomes.? Review of  
 
 Educational Research 72 (June 2002): 279-341. 
 
Betts, R.J. and Grogger, J. ?The Impact of grading standards on Student  
 
 Achievement,Educational Attainment and Entry-level Earngings.? Economics  
  
 of Education Review. 22 (2003): 343-352. 
 
Bowles, S. and Levin, M. H. ?The Determinants of Scholastic Achievement- An  
  
 Appraisal of Some Recent Evidence.? Journal of Human Resources.  
  
 3 (1968): 4-24. 
 
Chambers, G. J. ?The Impact of Collective Bargaining for Teachers on Resource 
  
 Allocation in Public School Districts.?  Journal of Urban Economics.  
  
 4 (1976): 324-339. 
 
Clarke, W.G.  History of Education in Alabama, 1702-1889.  Washington, Govt.  
 
 Print. Off., (1889):  269-295. 
 
Clotfelter, C.T. ?After Brown: the rise and retreat of school desegregation.?  
 
 Princeton University Press (2002): 1-50. 
 
 
 
 76
 
Cox, R. ?Kentucky?s Per Capita Personal Income: The Roles of Women and     
 
Education.? Kentucky Annual Economic Report. 70 (1997): 15-30.  
Decicca, P. P. and Lillard, R. L. ?Higher Standards, More Dropouts?  
Evidence Within and Across time.? Econ. of Ed. Rev.  20 (2000): 459-473. 
Domazlicky, B. ?Measuring the Cost of High School NonCompletion in                                             
 Southeast Missouri.? J. of  Econ. 22 (1996): 81-86. 
 
Durden, C. G. and Gaynor, E. D. ?More on the cost of Being Other than White and  
 
 Male: Measurement of Race, Ethnic, and Gender Effects on Yearly Earnings.?  
 
 American Journal of Economics and Sociology.  57 (1998):95-103. 
 
Ferguson, C. ?Public Education: A Crisis in Confidence.?  Alabama Business and   
 
 Economic Indicators. 68 (1999): 1-4. 
 
Figlio, N. D. and Stone, A. J. ?Are Private Schools Really Better??  Research in  
 
LaborEcon. 18 (1999): 115-140. 
 
Figlio, D. and Stone, J. ?Can Public Policy Affect Private School.? U.S. Department  
 
of Education, Office of Educational Research and Improvement. (2002): 4-40. 
 
Flower, M. R. ?Tuition Tax Credit and the Public Schools? National Tax Journal. 
  
41 (1988): 87-96. 
 
Freshwater, D. and Goetz, J.S. ?Social and Institutional Factors as determinants of  
  
 Economic Growth: Evidence from the United States Counties.? Papers Reg.  
 
 Science. 81 (2001): 140-155. 
 
Fullerton, T. ?Educational Attainment and Income Performance.? Econ. And Amp;  
 
 Finance Rev. 10 (2001): 2-10. 
 
 
 
 
 77
 
Greene, M. and Hoffnar, E. ?Residential Location and the Earnings of White and  
 
 Hispanic Men.? Applied Economics Letters. 1 (1994): 127-131. 
 
Hamilton, W.B. and Macauley, K.M. ?Determinants and Consequences of the  
  
 Private-Public School Choice.? Journal of Urban Economics. 29 (1991):  
  
 282-294. 
 
Hanushek, E. ?The Economics of Schooling: Production and Efficiency in Public  
  
Schools.? J. Econ. Lit. 24 (1986): 1141-1177. 
 
Hanushek, E. ?What If There Are No ?Best Practices.?? Journal of Political Economy.  
  
51 (2004): 156-172. 
 
Hanushek, E. ?Black-White Achievement Differences and Governmental  
 
Interventions.?  American Economic Review.  91 (2001): 24-28. 
 
Lankford, H. and Wyckoff, J. ?Primary and Secondary School Choice among Public  
 
and Religious Alternatives.? Economics of Education Review. 11 (1992):  
 
317-337. 
 
Long, E. J. and Toma, F. E. ?The Determinants of Private School Attendance.?  Rev.  
 
ofEcon. And Stat.  70 (1988): 351-357. 
 
Murphy, L.J. and Ribich, I.T. ?The Economics Returns to Increased Educational  
  
Spending.? The Journal of Human Resources. 10 (1975):56-77. 
 
Neal, D. ?What Have We Learned about the Benefits of Private Schooling??   
 
Economic Policy Review.  110 (1998): 79-86. 
 
Patrinos, A.H. ?Differences in Education and Earnings across Ethnic Groups in  
  
Guatemala.? The Quarterly Review of Economics and Finance.  37 (1997): 
  
809-821. 
 
 
 78
 
Rhine, S.L.W. ?The Effect of State Mandates on Student Performance.? American  
 
Economic Review. 79 (1989): 231-35. 
 
Rupert, P. and Schweitzer, M.E. ?Earnings, Education, and Experience.? Economic  
 
Review. 32 (1996): 2-20. 
 
Sonstelie, J.  ?The Welfare Cost of Free Public School.? Journal of Political   
 
Economy. 90 (1982): 794-808.  
  
Sonstelie, J. ?Public School Quality and Private School Enrollments.? National Tax  
  
 Journal. (1980)   :343-353. 
 
Tamura, R. ?Teachers, Growth, and Convergence.? Journal of Political Economy.    
  
 109 (2001): 1021-1059. 
 
Weiss, D.R. ?The Effect of Education on the Earnings of Blacks and Whites.? The  
  
 Review of Economics and Statistics. 52 (1970): 150-159.