Download as pdf or txt
Download as pdf or txt
You are on page 1of 22

Board of Regents of the University of Wisconsin System

Determinants of the Quality of Primary and Secondary Public Education in West Virginia
Author(s): Richard Raymond
Source: The Journal of Human Resources, Vol. 3, No. 4 (Autumn, 1968), pp. 450-470
Published by: University of Wisconsin Press
Stable URL: http://www.jstor.org/stable/144797
Accessed: 23-12-2015 23:29 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

Board of Regents of the University of Wisconsin System and University of Wisconsin Press are collaborating with
JSTOR to digitize, preserve and extend access to The Journal of Human Resources.

http://www.jstor.org

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
DETERMINANTS OF THE QUALITY OF
PRIMARY AND SECONDARY PUBLIC
EDUCATION IN WEST VIRGINIA*

RICHARD RAYMOND

ABSTRACT

Studiesanalyzingexpendituresfor public educationhave used a variety of


inputs into the educationalprocess as proxies for the quality of education.
This study attemptsto isolate some of the inputs which do, in fact, have
an effect upon educational quality. To accomplish this, output measures
of quality were derived from a sample of 5,000 West Virginia University
students who had graduated from high schools within the state. The
results show that only one of the input variables examined, teachers'
salaries, was significantlyrelated to the output measures of quality. In
addition, factors exogenous to the local school system-factors reflecting
socioeconomic characteristicsof the communities in which the school
systems were located-were observed to have a significant effect upon
quality. The results give rise to dual conclusions. First, input variables
seem to be imprecise measures of educational quality. Second, the
empirical evidence provides some support for the contention that the
quality of education may be improved by offering higher salaries to
teachers.

INTRODUCTION

The purpose of this study is to investigate the relationship between certain


inputs into the primary and secondary educational process and the quality
of education produced. Studies attempting to explain variations in spending
upon education often use inputs into an educational system as a proxy for

The author is Associate Professor of Economics, West Virginia University.


* The research upon which this paper is based was jointly supported by the
Regional Research Institute and the Human Resources Research Institute of
West Virginia University. Helpful suggestions were received from William
Miemyk, Roy W. Bahl, and Robert J. Saunders. The author assumes full
responsibility for the contents of the paper. An expanded version of this paper,
containing detailed descriptions of estimation procedures, is available upon
request from the Regional Research Institute of West Virginia Universty.
The Journal of Human Resources * III - 4

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
Raymond 1 451
the quality of output produced by the system.' Little attention has been
given to supporting arguments.
Whileit is doubtlesstrue that betteror more abundantinputswill lead
to a better quality output, this conclusion provides only weak support for
the use of a specific relationship between inputs and output quality. Empiri-
cal studies seem implicitly to assume that the relationship between inputs
and output quality is directly proportional, i.e., that a 10 percent increase
in teachers'salarieswill produce a 10 percent increase in the quality of
education.2 This may in fact be the case, but there is no a priori reason for
expecting this or any other specific relationship to hold. Some light may be
shed upon this point by investigating the relationship between output
measures of the quality of education and a number of specifically defined
input variables.
In this paper no attempt has been made to develop a comprehensive
measure of the. output of an educationalsystem. Instead it concentrates
upon a single aspect of this output: the adequacy of preparation for higher
education. A significant and growing percent of high school graduates attend
institutions of higher learning. The caliber of their preparation is thus
becoming quantitatively more important. In addition, this aspect of educa-
tion may prove representative of education in general. Admittedly, con-
centrationupon a single facet of the quality of education will limit the
applicability of the results. Suggestions for extending the scope of the study
are presented in the concluding section.
The empirical results were derived from data pertaining to the county
school districts in the state of West Virginia. These districts should be
relatively homogeneous with respect to most of the variables used in the
study.
The two quality measures used are described and a number of factors
often assumed to affect the quality of education are introduced with explan-
atory comments where necessary. Finally, the results of a multiple regres-
sion analysis are presented and interpreted.

1 See for example, Werner Z. Hirsch, "Determinantsof Public Education Expen-


ditures," National Tax Journal, 13 (March 1960), and John C. Bollens,
Exploring the Metropolitan Community (Berkeley and Los Angeles: Univer-
sity of California Press, 1961), pp. 324-26. In both cases the quality of
education is measured by an index composed primarily of input variables.
Statistically, this method proves quite successful. The "quality" variables are
significantly correlated with expenditures. But this is not at all surprising since
the "quality"variable contains some measure of teachers' salaries which are a
significant part of the expenditure being "explained."
2 Any study that includes teachers' salary as a proxy for quality is doing precisely
this. The index approach utilizing a combination of inputs involves the same
assumption in relation to the index itself rather than to its individual
components.

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
452 1 THE JOURNAL OF HUMAN RESOURCES

QUALITY MEASURES

Achievement test scores (ACT) and freshman grade point averages


(FGPA) were obtained for approximately 5,000 students entering West
Virginia University (WVU) between September 1963 and September
1966.3 These students, grouped by county of residence, make up the sample
from which the quality variables were derived.
The use of simple county averages for both test scores and freshman
grades would be appropriate if the sample of students from each county
was representative of the entire college-bound student population in the
county. But WVU does not obtain a random sample of college-bound stu-
dents from each county. Instead, the university seems to attract the best
from some counties and a decidedly poorer group from others.4 This
obviously renders invalid a comparison of educational quality among
counties based upon raw test scores and freshman grades pertaining to the
individuals in the sample. It was thus necessary to adjust the raw figures
to eliminate the inherent bias contained in the sample.
To accomplish this, senior year grades in selected high school subject
areas were used as a proxy for student aptitude. It was then assumed that
the marginal contribution of aptitude to test scores and freshman grades
was invariant among school systems differing in quality.5 Furthermore, a
linear relationship between aptitude and both ACT and FGPA was postu-
lated. The assumed relationships take the following form:

3 All students entering WVU are required to submit American College Testing
Program (ACT) test scores. The scores are broken down into five categories:
(1) English, (2) Mathematics, (3) Social Studies, (4) Natural Science, and
(5) Over-all. Only the over-all scores will be used in the current study. The
use of both ACT scores and FGPA was deemed advisable for two reasons.
First, it is possible that the ACT score gives a more comprehensive indication
of the quality of education received. The FGPA, however, must be considered
a better measure of the adequacy of preparation for higher education. Second,
ACT and other achievement test scores are often used as a measure of the
adequacy of preparation for higher education. By including both ACT scores
and FGPA's, it may be possible to determine if serious errors are likely to
result from this procedure.
4 This contention is based upon the range in the high school grade point average
(HGPA) by county within the sample. The highest average was 3.50 and the
lowest 2.49. These averages refer only to senior grades in English, Mathematics,
Social Studies, and Natural Sciences.
5 It is, of course, possible that the marginal contribution of aptitude is a function
of the quality of the school system, e.g., it may increase as the quality of the
school increases. Checking upon this was impossible since there are no a priori
grounds for classifying schools on the basis of quality.

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
Raymond 1 453

(1) ACT = a + b X HGPA


(2) FGPA =a + b' X HGPA
where b and b' are the marginal contributions of aptitude to ACT score
and FGPA, respectively. The quality of education received by an individual
is reflected through the value of the constants a and a'.
Values for b and b' were estimated by taking a 10 percent stratified
random sample from the 5,000 observations and regressing both ACT score
and FGPA upon HGPA. The regression equations obtained are (all GPA's
are measured on a 400.0 basis):
(3) ACT = 13.43 + .0300 HGPA
(4) FGPA = 31.23 + .6457 HGPA
Both regression coefficients are significant at the 1 percent level. Given
these values for b and b', test scores and freshman grades may be adjusted
to reflect only differences in the quality of education received. It is only
necessary to solve (1) and (2) for a and a'.
(5) a = ACT - .0300 HGPA
(6) a = FGPA - .6457 HGPA
The a's and dt 's so computed and aggregated over each of the 55 individual
counties served as quality variables.6

DETERMINANTS OF THE QUALITY OF EDUCATION

Quality of Instruction
There should be a positive relationship between teachers' salary and the
quality of education. If teachers are relatively mobile, then school systems
paying the highest salaries should procure the best teachers and thereby
offer the best instruction.7 In addition, high teachers' salaries may indicate

6 This procedure, in effect, equates the measures of quality to deviations from


the regression line used to estimate b and b'. In terms of the ACT measure,
the quality variable for an individual is equal to his actual score minus his pre-
dicted score plus the score intercept of the regression line used to estimate b.
7 Some investigators argue that mobility among teachers is insignificant. Jerry
Miner, in "Social and Economic Factors in Spending for Public Education,"
The Economics and Politics of Public Education (Syracuse, N. Y.: Syracuse
University Press, 1963), p. 78, states that "most salary differences among
teachers appear to be the result of state and local financial circumstances
coupled with strong preferences by teachers for particular geographical,loca-
tions. An index of beginning teachers' salaries under these circumstances is a
measure of the cost of obtaining a homogeneous input in various school
systems." No supporting evidence is presented.

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
454 | THE JOURNAL OF HUMAN RESOURCES

an experienced teaching staff which should also lead to better quality


instruction.
Four separate variables are used to measure teachers' salary: (1)
average salary (AS) for all teachers in the county, (2) AS for elementary
teachers, (3) AS for secondary teachers, and (4) a weighted salary
schedule.8
Variables 1 and 4 may allow differentiation between the effects of
mobility and experience. The weighted salary schedule is not affected by
the average length of service of teachers whereas average salary is. Vari-
ables 2 and 3 are included to determine if relevant differences exist between
the early and later years in the pre-college educational process.
The salary paid by a school system is not the only factor capable of
affectingthe quality of teacher hired. The locational preferencesof indi-
viduals may be such that teachers' salaries in near-by counties will have a
greater effect upon migration than will salaries in distant counties. This
possibility is investigated by including as an independent variable the
weighted average teachers' salary for all contiguous counties.9
The existence of attractive opportunities for employment outside the
teaching profession might also prove important. In terms of the available
data, it may safely be concluded that teachers will move primarily to the
Professional, Managerial, and Kindred occupations. Their movement should
be more rapid where salaries are high and jobs are plentiful in these non-
teaching occupations. Statistically, this effect is investigated through a
variable consisting of the median income for this occupational category
multiplied by the percent of total county employment found in the given
occupations.'0

Other Direct Determinants


There are many factors in addition to teachers' salaries which are capable
of directly affecting the quality of education. These factors are represented

8 This variable is a weighted average of the starting salaries for holders of Bach-
elor's and Master's degrees, adjusted for differences in average annual incre-
ments. The weights used are the numbers of teachers in the state with Bachelor's
and Master's degrees in 1960-61.
9 Contiguous counties in surrounding states are included in this variable.
10 A single variable has been used incorporating both the salary level and the
relative importance of Professional, Managerial, and Kindred occupations in the
county's economy, since the use of either factor separately could prove mis-
leading. For instance, high salaries would have little effect if very few job
opportunities were available. Unfortunately, the necessary data were available
only for male occupational categories. This is a serious shortcoming since the
majority of teachers in the state are female.

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
Raymond 1 455
by four variables: (1) the student-teacher ratio, (2) the percent of teach-
ers teaching in two or more fields, (3) the number of library volumes in
excess of standard, and (4) adjusted current expenditures per pupil.11
Variables 1 and 2 are measures of the demands (load) placed upon
the individual teacher. They should be negatively related to quality. Vari-
ables 3 and 4 are proxies for the adequacy of auxiliary facilities and thus
should be positively related to quality.12

Population Characteristics
The method used to derive the quality variables assumed that the aptitude
of students is roughly equivalent among counties. If this assumption is
incorrect, then some part of the intercounty differences in quality, as it has
been measured, will be attributable to aptitude differences among students.
A "quality" difference generated in this manner does not indicate a differ-
ence in the efficiency of school systems. Two identical school systems will
graduate a different quality of student if their student bodies differ notice-
ably in aptitude. The same type of problem occurs in relation to the
motivation of students.
Three variables are included as proxies for aptitude and motivation:
(1) percent of county employment in the Professional, Technical, and
Kindred category, (2) median family income, and (3) median school years
completed by adults. These variables should be positively related to the
quality measures.
A fourth variable,urbanization,was groupedwith the three proxies
for aptitude and motivationfor statisticalpurposes because it is highly
correlated with these three variables.'3 But the rationale underlying the
urbanizationvariableis different;it was includedto approximatethe degree
of school consolidation within counties. If consolidation is positively related
to the quality of education, expenditures being held constant, then urban-
ization should be positively related to quality.
11 The measure of current expenditures per pupil is somewhat unconventional.
To preserve the independence of teachers' salary, all payments to teachers have
been removed from the expenditure figures.
12 The regressions will contain no measure of the value or adequacy of school
plant. Reported figures on the value of school buildings, land, and equipment
are nearly useless. See the Bureau of Educational Field Services, Miami Uni-
versity, Educational Statistics in West Virginia (Oxford, O.: December 1960),
p. 43. There are, of course, other alternatives, but none adequately handles
the difficulties posed by the sporadic nature of both current capital expendi-
tures and innovations in the design of school plant. Any significant results
associated with an admittedly inadequate measure would have to be regarded
as statistical artifacts.
13 The measure of urbanization is simply the percent of total county population
living in cities and towns.

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
456 | THE JOURNAL OF HUMAN RESOURCES

Summary of Variables and Lagged Relationships


Before presenting a complete list of variables, the problem of lagged rela-
tionships should at least be mentioned. One description of the lags involved
is given by Mort: "These studies indicate that the effect of an expenditure-
level quality change is measurable even after twenty-five years, and that the
maximum impact occurs in about seven years."14 Thus, the quality of
education existing at a point in time is clearly a result of determinant levels
which prevailed in a number of past years.
The lag structure involved may prove to be quite complex. Lags will
undoubtedly differ among determinants, and they may also differ with
respect to the particular aspect of school quality being investigated. No
attempt has been made to incorporate a complex lag structure into the
analysis since it was impossible to obtain data covering a long time span
for over half of the independent variables.
An attempt was made to measure the independent variables at 1960
levels, although this was not always possible. Three considerations dictated
this choice: (1) a number of variables required data available only in the
1960 Census; (2) three of the four salary variables were unavailable prior
to the 1960-61 school year; and (3) the quality measures related to the
1963-66 period thus making the choice of 1960 conform roughly to
Furno's "maximum impact" period.
The following is a complete list of the variables used in the regression
analysis. The year, or years, to which each variable pertains is contained
in brackets following the description of the variable.15

14 Paul Mort, Walter Reusser, and John Polley, Public School Finance (New
York: McGraw-Hill Book Co., 1960), p. 86. The quote refers to a study by
Orlando Furno, "The Projection of School Quality From Expenditure Level,"
unpublished Doctoral dissertation, Teachers College, Columbia University,
1956.
15 Basic sources are as follows: Quality-see text. Salary-(1) State of West
Virginia, Department of Education, 48th Report of the State Superintendent
of Free Schools of The State of West Virginia; (2) West Virginia Educational
Association, Salary Schedules for Classroom Teachers in West Virginia (Octo-
ber 1961); (3) State Board of School Finance (West Virginia), The School
Dollar, a mimeograph report. Direct Determinants-(1) State of West Virginia,
Department of Education, 45th, 46th, 47th, 48th, and 49th Report of the State
Superintendent of Free Schools; (2) State Department of Education, West
Virginia Education Bulletin, Special Statistical Number (July, August 1961).
Population Characteristics-( 1) U.S. Bureau of The Census, U.S. Census of
Population 1960: General Social and Economic Characteristics, Vol. 50C; (2)
State of West Virginia, Department of Education, 47th and 48th Report of the
State Superintendent of Free Schools; (3) The urbanization measure was de-
rived from estimates used by Roy W. Bahl and Robert J. Saunders in Inter-
county Differences in West Virginia Government Expenditures (West Virginia
University, Business and Economic Studies, Vol. 10, No. 3, June 1967).

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
Raymond 1 457

QUALITY

Y1= Freshman grade point average quality measure, FGPA (1963-66)


Y2 =ACT test score quality measure,ACT (1963-66)

SALARY

X1 =Weighted salary schedule, WSS (1960-61 school year)


X2 =Average salary for all teachers,AS (1960-61 school year)
X3 =Average salary for elementary teachers, ELAS (1962-63 school
year)
X4 =Average salary for secondaryteachers, SCAS (1962-63 school year)
X5 =Contiguous county weightedsalaryschedule,CCWS(1960-61 school
year)
X6 =Contiguous county averagesalary, CCAS (1960-61 school year)

DIRECT DETERMINANTS

X7 =Percent of teachers teaching in two or more fields, %TT (1965-66


school year)
X8 =Student-teacherratio, STR (average for school years from 1957-58
through 1961-62)
X9 =Number of library volumes in excess of standard,LIBV (1960-61
school year)
X10= Currentexpenditureper pupil, CEPP (averagefor school years from
1957-58 through 1961-62)

POPULATION CHARACTERISTICS

X11=Alternativeemploymentopportunity,AEO (1960)
Xl2=Median family income, MFI (1959)
X13=Median school years completedby adults, MSYC (1960)
X14=Urbanization,URB (averageof 1957 and 1962)
Xl5=Percent employed in Professional, Technical, and Kindred occupa-
tions, %PTK (1960)

DESCRIPTION OF EMPIRICAL RESULTS

Table 1 gives simple correlation coefficients for all variables included in


the statistical analysis.16 Due to the many strong relationships between
16 These coefficients were derived from data pertaining to 49 of West Virginia's
55 counties. Gilmer, Lincoln, Pendleton, Summers, Wayne, and Wirt counties
were deleted from the sample because less than ten observations were avail-
able in each of these counties. Eight additional counties had less than 20
observations. Preliminary regressions showed that deleting these eight additional
counties produced no significant change in results. These counties were, there-
fore, kept in the sample.

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
458 | THE JOURNAL OF HUMAN RESOURCES

TABLE 1
SIMPLE CORRELATIONS OF ALL INDEPENDENT VARIABLES (Xl-X15)
AND TWO DEPENDENT VARIABLES (Y1 AND Y2)

Y1 Y2 X1 X2 X3 X4 X5 X6
FGPA ACT WSS AS ELAS SCAS CCWS CCAS
Y1
FGPA 1.0
Y2
ACT .8323 1.0
X1
WSS .3284 .5055 1.0
X2
AS .5460 .6051 .8780 1.0
X3
ELAS .5326 .5724 .8135 .9480 1.0
X4
SCAS .3031 .4204 .8973 .8962 .8874 1.0
X5
CCWS .3242 .4305 .6747 .6154 .5791 .6556 1.0
X6
CCAS .3032 .4025 .5475 .5390 .5153 .5708 .9615 1.0
X7
%TT -.2120 -.2474 -.4008 -.4687 -.3861 -.3997 -.1137 -.0558
X8
STR .1092 .2709 .5064 .3521 .3460 .3799 .3166 .2506
X9
LIBV .2782 .2106 .3091 .3853 .3957 .3287 -.0347 -.1393
X10
CEPP .2690 .1920 .3169 .5227 .4917 .4939 .2504 .2467
X11
AEO .4216 .4296 .4836 .623 1 .6004 .4973 .2352 .1567
X12
MFI .3487 .4470 .8224 .8479 .8223 .8505 .5388 .4658
X13
MSYC .4345 .3991 .4375 .6806 .7012 .5756 .2667 .2564
x14
URB- .3520 .3261 .4299 .6849 .6879 .6153 .3534 .3475
X15
%PTK .3943 .3559 .2939 .4221 .3956 .2653 .1762 .1403

independent variables, a careful survey of this table is a necessary pre-


requisite to interpretation of the regression results. A sketchy summary of

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
Raymond 1 459

TABLE 1 (Continued)

X7 Xs X9 X1o X11 X12 X13 X14 X15


%TT STR LIBV CEPP AEO MFI MSYC URB %PTK

1.0

-.3428 1.0

-.2974 .1420 1.0

-.1442 -.2753 .3588 1.0

-.5491 .2115 .5998 .3387 1.0

-.6411 .4395 .3290 .3823 .6726 1.0

-.5479 .1639 .4560 .4513 .7776 .7581 1.0

-.4523 .0878 .3265 .4761 .6865 .6851 .8337 1.0

-.3463 .0557 .4168 .1861 .8164 .4479 .5892 .4527 1.0

these intercorrelations might stress the following points: (1) the popula-
tion characteristic variables are strongly related both to one another and

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
460 | THE JOURNAL OF HUMAN RESOURCES

to the salary variables; (2) the contiguous county salary variables bear a
moderate to strong relationship to the salary variables pertaining to the
county itself; and (3) the direct determinants seem to be only weakly
related to all of the other variables.
Table 2 presents the partial correlation coefficients and R squares
obtained from the regression analysis. The two quality measures were used
as dependent variables, and X7 through X15Q were included as independent
variables in all regressions. In addition each regression contained one salary
variable and the corresponding contiguous county salary variable. The
amount of explained variation in quality ranged from over 47 percent in
regression 6 to under 30 percent in regressions 1 and 4. In all cases, the
same set of 11 variables explained more of the variation in test scores than
in freshman grades.
The only significant relationships are found among the salary vari-
ables. Differences among the salary variables are essentially the same as
those observed in the simple correlation matrix; average salary for ele-
mentary teachers is significant in both regressions, whereas average salary
for secondary teachers is completely unrelated to quality when other fac-
tors are controlled.
Holding other factors constant does not render significant the shift
parameters on the supply curve of teachers (X5j,X6, X11). With one excep-
tion, these variables continue to carry positive signs. But the high simple
correlations between these variables and the salary measures make it
impossible to rule out misleading results due to collinearity.
The direct determinants (X7, X8, Xg, Xlo) bear absolutely no rela-
tionship to either quality variable when the level of other variables is held
constant. An examination of the simple correlation matrix makes it appear
very unlikely that this is a result of collinearity. The uncontrolled relation-
ships between the direct determinants and quality were quite weak. The
direct determinants were related to the salary and population characteristic
variables, but the relationships in general were not very strong.
The population characteristics present a confusing and conflicting
picture. The negative signs carried by family income (X12) and urbaniza-
tion (X14) must be attributed to the strong relationships between these
variables on the one hand and the salary measures and adult educational
achievement (AX3) on the other. Furthermore, the simple correlations
among the population characteristic group are uniformly so high that no
significance can be attached to their partial correlation coefficients. The
regressions give no reliable information about the role of the population
characteristics.
In addition,variableswere added to each of these regressions,one
at a time beginning with the strongest, until all variables carrying a signifi-

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
Raymond 1 461

cant regression coefficient had been included.17 It was impossible to find


more than one significant variable for any of the eight regressions. The
significant variable in each case was, of course, the variable among the
given set of independents showing the highest simple correlation with the
dependent variable (see Table 1). The average percent increase in R
square from the 1 variable to the 11 variable regressions was 46.47. In
other words, on the average a single variable explained approximately
two-thirds as much of the variation in quality as did 11 variables.
Since alternative employment opportunity (Xi,) and the population
characteristics were so closely interrelated, an index was formed covering
all five of the population characteristic variables. A principal component
analysis was used to derive the index.18 The strongest component extracted
from the five variables explained 74 percent of the variance among the
variables. The simple correlations between this component and the five
variables were: AEO = + .9195, MFI = -+ .8309, MSYC = + .9265,
URB = + .8564, and %PTK = + .7601. Thus the component not only
explains a large portion of the variation among the variables, but it seems
to be a good proxy for each of them as well. This component, denoted by
P, may consequently be used as an index of general population character-
istics.19 Simple correlation coefficients involving this index are presented in
Table 3.
Regressions 1 through 8 were recalculated using the index P in place
of variables X12, X13, X14, and X15, e.g., the new regression 1, henceforth
denoted by 1', containedindependentvariablesX1, X5, X7, X8, X9, X0o,
and P. The coefficients of determination were lower in all regressions
when only these seven variables were included. They ranged from .2711 in
regression 4' to .3999 in regression 6'. Higher levels of significance were,
however, associated with the lower R squares. The variables again ex-
plained a larger portion of the variation in test scores than in freshman
grades, and significant relationships were again observed primarily among
the salary variables. The sole exception was regression 4' in which P was

17 Since the direction in which each of the variables should act was specified, a
one-tail test was used. The significance level was set at 5 percent.
18 For a general description of principal component analysis, see M. G. Kendall,
A Course in Multivariate Analysis (London: Charles Giffen and Company
Limited, 1957), ch. 2, or Donald F. Morrison, Multivariate Statistical Methods
(New York: McGraw-Hill Book Co., 1957), ch. 7.
19 In arriving at this index, a component representing only the population char-
acteristic measures was first derived. The component was adequate, but it was
correlated with alternative employment opportunity (AEO) at approximately
.9. Since this component and AEO obviously could not be used in the same
regression, AEO was added to the other four variables and the component to
be used was derived.

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
TABLE 2
PARTIAL CORRELATION COEFFICIENTS DERIVED FROM ALL REGRESSIONS CONTAINING
INDEPENDENT VARIABLES

Independent Variable
Regression Dependent XT X8 X9 Xo1 Xli X12
Number Variable %TT STR LIBV CEPP AEO MF
1 FGPA-Yi -.0564 .0227 .0354 .0666 -.0026 -.1757
2 FGPA-Y1 -.0686 -.0561 .0056 -.1017 .0375 -.3547
3 FGPA-Y1 -.1569 -.0361 -.0019 -.0175 .0461 -.3159
4 FGPA-Y, -.0191 .0482 .0938 .0803 .0224 -.0546
5 ACT-Y2 -.0397 .0539 -.0894 .0182 .0900 -.2041
6 ACT-Y2 -.0224 .0023 -.0850 -.1470 .1592 - .263
7 ACT-Y2 -.1050 .0270 -.0804 -.0565 .1600 -.2013
8 ACT-Y2 .0013 .0893 -.0035 .0148 .1352 -.0328
,. .

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
TABLE 2

Independent Variables

Regression Dependent X1 X2 X3 X4 XZ X6
Number Variable WSS AS ELAS SCAS CCWS CCA
1 FGPA-Y1 .1151 .1923

2 FGPA-Y1 .4970* .12


3 FGPA-Y1 .4193* .17
4 FGPA-Y1 -.0301 .243

5 ACT-IY' .2597* .1562

6 ACT-Y2 .4752* .17


7 ACT-Y2 .3895* .21
8 ACT-Y2 .0520 .258

* Significantat the 5 percent level with a one-tailed test.


a The hypothesis associated with this variable specified a positive relationship;thereforeth
b Variablesdeleted to renderR
square significant.

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
464 J THE JOURNAL OF HUMAN RESOURCES

the only significant variable. This is not, however, sufficient reason for
concluding that only salaries are related to quality. The pattern exhibited
by the salary variables and P indicate that collinearity affected the results.
There were three cases, regressions 1', 4', and 8', in which P showed a
higher simple correlation with the quality variable than did the salary
variable included in the regression. In all three cases the full weight of
the regression fell on P, leaving the salary variables not only insignificant
but also carrying a negative sign. There were four cases, regressions 2',
3', 6', and 7', in which the salary variables included in the regression
showed a higher simple correlation with the quality variable than did P.
In all four cases the salary variables carried the full weight of the
regression and P was completely insignificant. These regressions did not
appear reliable with respect to the independent effects of P and the salary
variables.
The results did, however, tend to corroborate previous observations
in relation to the direct determinants. None of the determinants exhibited
significant associations with quality. Collinearity did not appear to be
important in this case.
As in the case of regressions 1 through 8, little was accomplished
by adding variables to the seven variable regressions one at a time until
all variables carrying a significant regression coefficient had been in-
cluded. Only one variable proved significant for all eight regressions.20
The average percent increase in the coefficient of determination from 1
variable to the 7 variable regressions was 20.55, i.e., one variable ex-
plained over 80 percent of the variation in quality which was associated
with all seven variables together.2'
The results clearly indicate that the four direct determinants included
in the study did not have a significant effect upon the quality of education.
Two of these measures, the percent of teachers teaching in two or more
fields and the adequacy of library facilities, deserve little comment. They
simply did not prove as important as might have been expected. In addi-
tion, it should be rememberedthat the measureused for currentexpendi-
ture per pupil was not a conventional one. Total person service expense
for instruction was deleted so that the effect of teachers' salaries could be
investigatedseparately.If this item had not been deleted, it is quite prob-
able that current expenditures would have proven significant. Such a
result, however, would reflect primarily the effect of salary differences.

20 A one-tail test with the significance level set at 5 percent was used. Regression
8' produced two variables, P and X6, with t values greater than 2.0. X6 was
judged insignificant since it carried a positive sign.
21 An attempt to find a single principal component covering the four direct
determinantswas unsuccessful.

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
Raymond 1 465

TABLE 3
SIMPLE CORRELATION BETWEEN THE POPULATION CHARACTERISTICS
INDEX (P) AND ALL OTHER VARIABLES

Variable R
Y1 -FGPA .4536
Y2 -ACT .4550
X1 -WSS .5735
X2 -AS .7598
X3 -ELAS .7488
X4 -SCAS .6547
X5 -CCWS .3637
X6 -CCAS .3166
X7 -%TT -.5926
X8 -STR .2238
Xg -LIBV .4976
XIo -CEPP .4308

The performance of the student-teacher ratio may have resulted from


the specific nature of the educational systems in some of the West Virginia
counties making up the sample. Although none of the correlations provide
supporting evidence, a survey of the data suggests that a number of
low student-teacher ratios may have been caused by underutilization of
teachers in some of the more rural counties. In these instances, the low
student-teacher ratio serves as a proxy for an inadequately consolidated
school system and as such should be associated with low quality.
The data also fail to support the contention that a negative relationship
exists between the quality of education in a given county and salaries
in immediately surrounding counties. Statistically, the high correlations
between the contiguous county salary variables and the salary variables
prevent us from obtaining reliable estimates of the independent effects of
each. It is, however, safe to conclude that if high salaries in surrounding
counties do negatively affect the quality of education in a given county,
then this effect is overridden by the positive effect of the high salaries
generally paid by the county in question.
The contiguous county salary variables were included because it was
deemed likely that salary differences among nearby counties would provide
a greater stimulus to migration than would salary differences among
widely separated counties. While the results do not disprove this conten-
tion, they do suggest that the migration most important in the actual
determination of educational quality does not occur among contiguous

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
466 | THE JOURNAL OF HUMAN RESOURCES

counties. The similarity of salaries among contiguous counties apparently


makes movement beyond the borders of nearby counties necessary if an
individual is significantly to improve his salary status. Furthermore, with-
out claimingany statisticalsupportfor the position, it might be suggested
that the relatively strong relationship observed between the contiguous
county salary variables and the salary variables themselves could point
to a causal connection between the two. Similar demands among counties
for education, in terms of quality, coupled with a sufficient amount of
teacher mobility may force counties to keep their salaries roughly in line
with those paid in immediately surrounding areas.
Before discussing the population characteristic measures, the position
occupied by the alternative employment opportunity variable must be
clarified. The attempt to arrive at a measure for alternative employment
opportunities which would discriminate between this factor and certain
other independent variables was unsuccessful. This is not too surprising.
Alternative employment opportunities quite naturally are attractive where
income and education are high and where employment in professional
occupationsis important.The AEO variablein essence, if not in intent, is
an additional measure of the population characteristics assumed to affect
aptitudes and motivation. It will not be discussed separately.22
A discussion of individual effects among the remaining population
characteristic measures must rely almost exclusively upon the simple
correlation matrix. Collinearity obviously renders the regression results
unreliable. The task, however, is not as difficult as it may at first seem
since basically there is only a two-fold division among the four remaining
populationcharacteristicsvariables.First, it has been arguedthat where
family income, adult educational achievement, and professional employ-
ment are high, the aptitude and motivation of students will be high, thereby
producing higher values for the quality measures. Second, the contention
was made that, within our sample, high levels of urbanization represent
economies in the operation of school systems. Therefore, urbanization,
ceteris paribus, should be associated with high quality.
It is difficult to muster support for the urbanization hypothesis, but the
results do provide some support for the hypothesis associated with family
income, adult educational achievement, and professional employment.
Both the correlations involving the individual variables and those relating
to the index P show significant positive associations between the popula-
tion characteristicsand quality.While it is true that this may be a result

22 Since high values for AEO should lead to lower quality, its effect will oppose
the assumed effect of the other variables making up the index P. Thus the
independent positive effect of the remaining variables may be somewhat stronger
than indicated by the results.

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
Raymond j 467

of the high correlations between these variables and the salary measures,
it seems more reasonable to conclude that both salary and population
characteristics have an effect upon quality, but that the independent effect
in each case is somewhat weaker than shown by the appropriate simple
correlation coefficient.
This conclusionmakesit appearthat a partof the measureddifferences
in quality among counties is not attributable to differences in the quality
of education offered. The portion of the quality differences, as they have
been measured, which results from differences in population characteristics
falls largely outside the control of the school system. This portion is caused
by differences in student ability and home environment. It would be helpful
to remove this portion from the total and then to examine the determinants
of the remainder, but again collinearity renders this impossible.
The figures also support the contention that high teachers' salary will
lead to higher quality education. It is quite likely that the strength of this
relationship is overstated by the simple correlations between the salary
variables and quality (see above). But to argue that no significant effect
remains when other factors have been controlled seems clearly in conflict
with the data under consideration.
These results indicate that teacher mobility is an important factor in the
determination of the quality of education. This is somewhat surprising in
view of the fact that 73 percent of the teachers in the state were female.
As secondary breadwinners, many of these women will not move in
response to salary differentials. In addition, as previously pointed out,
salaries among contiguouscounties are quite similar.It might, therefore,
be concluded that salary differences affect quality by inducing a relatively
small proportion of teachers to migrate. Unfortunately, little additional
informationabout the migrant group is available. In terms of quantity,
the most important migration may relate to new college graduates ac-
cepting their initial positions.23 It is difficult even to speculate about the
qualitative characteristics of the migrants, although such information, if
available, would be quite useful.24

23 For present purposes such individuals would be classified as migrants if they


did not return to their original county of residence.
24 The use of high school grades as a proxy for aptitude may have introduced a
bias into the analysis. It is often assumed that higher grades are given in the
smaller rural schools. No data were available which would allow a clear
evaluation of this assumption. Within the sample, high school grades bear a
strong negative relationship to average salary (-.7042). Since the rural
counties in general pay lower salaries, it is safe to conclude that the high school
grades of rural students entering WVU are higher than the high school grades
of students entering from the more urban areas of the state. This, however,
does not establish that rural schools tend to give higher grades. If the rural

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
468 j THE JOURNAL OF HUMAN RESOURCES

The relationship between average salary and quality proved stronger


than that between the weighted salary schedule and quality. This may be
partially explained by the fact that average salary is significantly related
to the percent of county teachers holding a Master's degree (+.4080),
whereas the weighted salary schedule is not (+.0470). This, of course,
assumes that a teacher with a Master's degree is a superior teacher.
Neither the weighted salary schedule nor average salary bears a significant
positive relationship to the average length of service of teachers.
The marked difference in results between elementary average salary
and secondary average salary also deserves comment. The very high cor-
relations between the two variables prevents isolation of the independent
effects attributable to each. Elementary salary does, however, appear to
have a somewhat stronger effect upon quality than does secondary salary.
This is illustrated by the simple correlation coefficients between the two
salary variables and the quality measures.25 The corresponding regression
coefficients do not differ significantly from one another at the 5 percent
level, but elementary salary coefficients do exceed the secondary salary
coefficients in both cases, i.e., a given dollar change in elementary salaries
produces a greater change in quality than an equal dollar change in
secondary salaries.
These results are consistent with the belief that the elmentary level is
the most important stage in the pre-college educational process. They do
not, however, establish that quality differences among teachers are more
important at the elementary level. Elementary teachers may simply be
more mobile in response to salary differences than are secondary teachers.
Finally, the results associated with the two measures of quality are not
identical. The independent variables, both individually and collectively,
bear a closer relationship to the test score quality measure than to the
freshman grade quality measure. There is no readily apparent explanation
for this difference. It is, however, worth noting since it indicates that the
choice between these two quality measures is not a matter of indifference.
The results show that certain variables are significantly related to one
quality measure but not to the other.

counties send a larger portion of their best students to WVU than do the urban
areas, the figures may be consistent with similar grading standards throughout
the state. But if high school grades were undeservedly high in low salary
counties, a downward bias in the quality measures for these counties would
result. Given the estimation procedure used, this would cause an overstatement
of the positive relationship between teachers' salary and quality. With the
information available, it cannot be determined if this, in fact, did occur.
25 The difference between ELAS and SCAS is significant at the 1 percent level
in each case.

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
Raymond 1 469

SUMMARY AND CONCLUSION

Empirical results pertaining to the state of West Virginia give no support


to the use of certain input variables as proxies for the quality of educa-
tion. There was no evidence of a significant relationship between quality
and the student-teacher ratio, the percent of teachers teaching in two or
more fields, current expenditures, or the adequacy of library facilities,
Thus, in spite of obvious and perhaps convincing arguments in support
of these factors, it appears that, in fact, they are not always accurate
indicators of quality.
No such unequivocal statement relating to teaching salaries may be
made. The significance of the salary variables provides some justification
for the use of these variables as proxies for quality. On the other hand,
it should be noted that the highest coefficient of determination between
a salary variable and a quality variable was about .36. This, in conjunction
with the strong relationships between salaries and population character-
istics would seem to warrant only a very guarded use of salaries to
measure quality. The use of salary levels as a proxy for quality may give
rise to particularly difficult problems of interpretation when the aim of
the analysis is to explain variations in expenditures upon education (see
footnote 1).
But while it is certainly true that salaries are not a precise measure
of quality, it is perhaps more important to note that salaries do have some
effect upon quality. The exact nature of the relationship between salaries
and quality could not be determined with available data. It. may, however,
be concluded, in a qualitative sense, that one way to better education is
through higher teachers' salaries. This is a very important "fact"-one
that has long been appreciated by educators but not by taxpayers.
Some concluding comments may be made upon the large portion of
unexplained variation. No combination of the independent variables was
able to explain as much as 50 percent of the variation in quality. There
are many possible explanations for these rather poor results. The quality
measures were far from perfect. This is also true of some of the estimation
procedures necessitated by data inadequacies. In addition, there are defi-
nitely important factors which were not included, e.g., adequacy of school
plant. It is unlikely, however, that these difficulties are primarily respon-
sible for the nature of the results.
A more serious problem relates to the objectives of the various school
systems. There is no reason to believe that all of the school systems within
the state devote the same portion of their effort and resources to the
preparation of college-bound students. Indeed, it would be a malallocation
of resources if they did. The proportion of high school graduates going on

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions
470 | THE JOURNAL OF HUMAN RESOURCES

to college varies widely throughout the state, from 11 to 48 percent by


county in 1961. Many schools may place little emphasis on college pre-
paratory courses if year after year few of their students go on to college.
This will render difficultthe task of statisticallyexplainingquality as it
has been measured.
In addition, this raises the question of whetherit is generallyvalid
to lump together school systems with varying objectives in an investigation
directed more at one group of systems than at another in terms of these
objectives. This question actually relates to the proper methodology to
be followed in an attempt at comprehensive evaluation of the quality of
education and its determinants. This paper was not intended to be a com-
prehensive study of the quality of education. It purposely investigated
only a single aspect of quality. Furthermore,it may be that this is the
best way to begin an investigation of quality in general.
The separate aspects of the product of an education system are in
essence noncomparable, and attempts to render them comparable through
the construction of indices are fraught with difficulty. It may prove more
fruitful to investigate different aspects of quality separately and then to
attempt a consolidation of the results obtained from the separate studies.
In this form the suggestion is, of course, much too general to be useful.
Delineation of the quality areas to be studied presents a major problem
which for the empirical investigator would have to be solved on the basis
of the availability of concrete output measures.

This content downloaded from 138.73.1.36 on Wed, 23 Dec 2015 23:29:53 UTC
All use subject to JSTOR Terms and Conditions

You might also like