|
Predicting the Annual Salaries of Construction Educators using Multiple Regression
Richard
Burt and Ifte Choudhury
Texas
A&M University College
Station, Texas |
The development of a mathematical model to predict the annual salaries of construction educators is presented. A review of the literature identified a number of factors that are hypothesized to affect the annual salary of construction educators; academic qualifications, longevity, academic rank, parent college of the department, region in which the institution is located and gender. The responses from the annual ASC Faculty Salary Survey were used to develop a multiple regression model that predicts the annual 9-month salary of a construction educator. The stepwise selection method was used to select seven independent qualitative or dummy variables to include in the model. The model developed does not have a very high predictive efficacy as only 51 percent of the variation in the dependent variable (annual 9-month salary) is explained by the variation in the selected independent variables. The variables selected for the model includes levels of academic rank, academic qualifications, region in which the institution is located and parent college of the department. Independent variables representing longevity and gender were not included in the model. Key Words: Construction Educators, Salaries, Regression, Predictive Models |
Introduction
How
much should I be earning? It is a
question that most of us ask, but very few of us receive a satisfactory answer,
particularly those of us who teach in institutions of higher learning. Studies
on salary compensation are of interest to both academicians and administrators.
There exists a wide body of literature on faculty salary levels in institutions
of higher learning. One of the basic tenets of market economy is that income
should be distributed according to contribution, and none takes greater pride in
rewarding people for merit alone then academicians. Some studies seek to explain
the difference in salary levels in terms of performance and contributions. The
purpose of this study is to identify whether there are other factors that affect
faculty compensation, particularly those engaged in teaching in the member
schools of the Associated Schools of Construction.
Literature
indicates that apart from scholarly productivity, longevity makes a substantial
contribution to faculty salaries (Ferber, 1974; Monks & Robinson, 2001).
Studies suggest that when adequate measures of past mobility are controlled for,
the evidence of a positive correlation between income and seniority of academic
faculty is overwhelming.
There
is great dissatisfaction with the salary equity of women faculty in some
universities. Findings by Bellas et
al. (2001) show that a sizable gap between men and women’s salaries exists
after controlling for academic qualifications. The study indicates that
women’s scarcity in higher-level faculty positions contribute to their slower
promotion rates, which in turn depress their salary growth rates.
Faculty
salaries also differ across the disciplines. A study by Cox (2001) shows that
average earnings by law professors, at both public and private universities, are
the highest among all disciplines. Similar findings are reported by an earlier
study by Schneider (1999). Construction science is taught in schools ranging
from architecture to business to technology in different universities. It is
likely that the school in which they are employed may affect the salaries of the
faculty in the department of construction.
Most
of the studies on faculty salary include academic rank as an endogenous factor
for prediction of salaries. Webster (1995) reports that while salary of a full
professor differs from that of faculty in other ranks, the associate and
assistant professor ranks have no statistically significant relationship with
salary. He suggests that the outcome may be due to the effect of salary
compression.
In
view of the evidence provided by the results of the studies conducted in other
disciplines, it is hypothesized that faculty salary in schools of construction
that are members of the Associated Schools of Construction are affected by the
following factors:
Academic
qualifications | |
Longevity | |
Academic
rank | |
Parent
college of the department | |
Region
in which the institution is located | |
Gender |
Method
Study
Population
The
study population consists of all Faculty that teach at institutions that are
members of the Associated Schools of Construction (ASC).
The ASC web site identifies that there are 585 ASC Faculty.
Data
Collection
During
the fall of 2001 ASC faculty were sent email messages inviting them to take part
in the survey. The initial
invitation was sent out on the 11 October and follow up messages were sent on 18
and 28 October. The messages
invited faculty members to visit the ASC web site and complete an on-line
survey. A screen capture from the
web site showing the on-line survey form is shown in Figure 1.
By November 16, 2001, 200 of the 585 ASC Faculty had responded, a
response rate of 34.2 percent. From
the 200 respondents, 21 selected not to participate leaving 181 Faculty
completing the survey. Of the 181
Faculty completing the survey, 5 Faculty did not supply a full-time annual
salary figure and these responses therefore were not used in constructing the
statistical model. Permission was
sought and obtained from the ASC to use the database for the purposes of this
study. The ASC board had
previously voted to allow any program or faculty to have access to data within
the ASC Database. The
data provided has been cleaned so that it is not proprietary and does not
identify any particular individual.
|
Figure
1:
ASC Faculty Salary Survey Update form. |
Variables
of Interest
The
dependent variable of interest is the annual contract salary of ASC faculty for
a period of 9 months (ASC_9MO). Respondents
submitted their annual contract salary and the annual contract period in months.
Multiplying the annual contract salary by 9 and dividing it by the number
of months of the annual contract calculated the dependent variable. The predictor or independent variables are of two types:
quantitative and qualitative. The
quantitative predictor variables are:
Age
of the faculty (AGE) | |
Years
in current rank (YRS_RANK). |
The
qualitative or dummy variables are used to represent further information about
the faculty. The dummy variables
cover the qualifications, rank, geographical region, college and gender of the
faculty. A value of 1 is assigned
if a faculty is a member of a particular qualification, rank, geographical
region, college or gender group and a 0 if they are not. The qualitative or dummy variables are:
Baccalaureate
Degree (BS) | |
Master
of Science Degree (MS) | |
Master
of Arts Degree (MA) | |
Law
Doctorate (JD) | |
Doctor
of Education (DED) | |
Doctor
of Philosophy (PHD) | |
Assistant
Professor (ASST_PRO) | |
Associate
Professor (ASSO_PRO) | |
Full
Professor (FULL_PRO) | |
Department
Head or Chair (DEPT_HD) | |
Senior
Lecturer (SNR_LECT) | |
Lecturer
(LECTURER) | |
Visiting/Adjunct
Instructor (INSTRUCT) | |
Graduate
Teaching Assistant (GTA) | |
Far
West Region (FAR_WEST) | |
Great
Lakes Region (GT_LAKES) | |
North
Central (N_CENTRAL) | |
North
East Region (N_EAST) | |
Rocky
Mountain Region (ROCKY_MT) | |
South
Central Region (S_CENTRAL) | |
South
East Region (S_EAST) | |
Architecture
(ARCH) | |
Business
(BUSI) | |
Engineering
(ENGR) | |
Technology
(TECH) | |
Other
College (OTHER) | |
Male
(MALE) | |
Female
(FEMALE) |
The
names within parentheses are the names assigned to the variables for use in the
statistical software used for the analysis, SAS® for Windows®
version 8.
Hypothesis
The
hypothesis is that the 9-month annual salary of ASC Faculty can be predicted
using the following multiple regression model:
ASC_9MO
= β0 + β1AGE + β2YRS_RANK +
β3BS + β4MS + β5MA + β6JD
+ β7DED + β8PHD + β9ASST_PRO +
β10ASSO_PRO + β11FULL_PRO + β12DEPT_HD
+ β13SNR_LECT + β14LECTURER + β15INSTRUCT
+ β16GTA + β17FAR_WEST + β18GT_LAKES
+ β19N_CENTRAL + β20N_EAST + β21ROCKY_MT
+ β22S_CENTRAL + β23S_EAST + β24ARCH
+ β25BUSI + β26ENGR + β27TECH +
β28OTHER + β29MALE + β30FEMALE +
e. |
The
regression model shows that the dependent variable ASC_9MO
is a function of an intercept value (β0) and a series of
independent variables numbered from 1 to 30 and an error value. The “β” values are the parameter estimates and are
calculated using the statistical software.
These are multiplied by the setting of the independent variable to arrive
at a value for the dependent variable.
Development
of the Statistical Model
A
multiple regression model was developed to express the relationship between the
dependent variable and the independent variables. The regression model was developed using the three-step
approach set down by Ott (1993) namely; selecting the independent variables,
forming a suitable model and checking the model assumptions. The selection process identified those independent variables
that caused the greatest variation in the dependent variable.
The stepwise selection method was used to select the variables in the
model. The significance level
(P-value) for an independent variable to enter and remain in the model was set
at 0.10.
Results
of the Stepwise Selection Method
The
results of the multiple regression analysis are set out in Table 1.
The F value is used to test whether there is a regression relation
between the dependent variable and the independent variables.
The high F value and low P value (<0.0001) show that there is a
regression relation. This in itself
however does not mean that the model is suitable for predicting annual salaries. The R-Square value is the coefficient of determination and
measures how well the regression fits. The
R-Square value of 0.5140 shows that approximately 51 percent of the variation in
the 9-month annual salary is explained by the variation in the selected
independent variables.
Table
1 Results of the multiple regression procedure |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
The
independent variables selected by the multiple regression process and their
parameter estimates are set out in Table 2.
All variables selected for the model have a significance level of P <
0.10. All the variables selected
were qualitative or dummy variables, therefore the parameter estimate is the
amount in dollars that is added to or subtracted from the intercept value.
Table
2 Parameter
estimates for multiple regression model including 95 percent upper and lower
confidence intervals |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
Figure
2 is a chart showing predicted 9-month salary values with 95 percent confidence
and prediction intervals and actual 9-month salary values for all 176
observations. The values for
predicted and 95 percent confidence and prediction intervals were calculated
using the statistical software. For
example, the author is an Assistant Professor at a University in the
South-central region in a College of Architecture and has a Ph.D.
To predict his 9-month salary the following equation is used:
ASC_9MO
= 52605 + 4927*PHD + 7191*ASSO_PRO + 20092*FULL_PRO + 22322*DEPT_HD – 4765*N_CENTRAL
- 5927*S_CENTRAL – 4561*TECH + e. |
The
values for the independent variables PHD and S_CENTRAL are 1.
The settings for all other independent variables are 0.
The prediction therefore is:
ASC_9MO
= 52605 + 4927 – 5927 |
The
predicted 9-month salary is therefore $51605.
This is considerably more than the author’s actual salary of $46,800.
The author’s salary does however fit within the 95 percent lower and
upper prediction intervals that were calculated by the statistical software to
be $31,798 and $71,374 respectively.
|
Figure 2: Chart showing
predicted 9-month salary values and actual 9-month salary values |
In
light of the independent variables selected using the stepwise procedure, the
salary prediction model can be rewritten as follows:
ASC_9MO
= β0 + β8PHD + β10ASSO_PRO +
β11FULL_PRO + β12DEPT_HD + β19N_CENTRAL
+ β22S_CENTRAL + β27TECH + e. |
The
results show that faculty salary is correlated with one academic qualification
variable, three of the faculty rank variables, two geographical region
variables, and one college type variable. The regression model explains
approximately 51 percent of the variation in the 9-month salary.
This means that the predictive efficacy of the model is not very high.
This can be seen in Figure 2. The
model appears to predict 9-month salaries between the values of $45,000 and
$80,000 quite well, as most of the actual 9-month salary values lie within the
95 percent confidence intervals. At
the extreme ends of the salary scale the model is not so good at predicting
salaries, although the actual 9-month salary values lie within the 95 percent
prediction intervals. There is some
concern that some of the lower salary values may not be for full-time faculty
members. If this were the case then
it would be prudent to change the ASC Faculty Salary Survey Update form to allow
a faculty member to enter the percentage of a full-time salary they receive. In
fact, the literature indicates that studies on faculty salary are usually
conducted using full-time effort criterion (Monk & Robinson, 2001;Webster;
1995). The Information Management and Testing Services (1996) at Baylor
University defines a full-time faculty as a member
of the instructional staff who is employed full-time and whose regular
assignment is instruction, including those with released time for research.
The
model only selected PHD from the qualification dummy variables and suggests that
the possession of a PHD adds approximately $ 4927 to a faculty member’s
9-month salary. Evidence in the literature provides support for this finding
(Lamb & Moates, 1999; Webster, 1995).
Three
faculty rank dummy variables were selected; this was expected as most promotions
in academia result in salary increases. The model indicates the rank of
Associate Professor would increase the 9-month salary by $ 7191, the rank of
Full Professor by $ 20092, and the rank of Department Head by $ 22322. It will
be interesting to see in future studies whether these differences remain
significant if a productivity measure is introduced in the model.
Only
two of the seven geographical region dummy variables were selected.
The model indicates that Faculty in the North Central and South Central
regions would have their 9-month salaries reduced by $ 4765 and $ 5927
respectively. It will, however, be interesting to see whether these differences
remain significant after adjusting for variations in taxes and cost of living
across geographical locations.
The
only college dummy variable selected was Technology. The model indicates that being a faculty member in a College
of Technology would decrease their 9-month salary by $ 4561. There is some
evidence in the literature to support this finding. Schneider (1999) studied
differences in salaries of university professors by disciplines in four-year
institutions. The findings suggest that the average faculty salary in colleges
of technology is significantly lower than that in colleges of architecture,
business, and engineering.
The
variables that were not selected are also of interest. Neither of the two
quantitative independent variables, age and years in rank, was found to be
statistically significant. Even though the general body of literature suggests
that a correlation exists between salary compensation and these two variables, a
study by Moore et al. (1998) provides evidence contrary to this belief. The
researchers did not find any positive relationship between either longevity or
seniority with faculty salary when the model took into account only these two
independent variables. The variables, however, became significant when a
quality-adjusted measure of research publications was introduced in the model.
It will be worthwhile to introduce such a variable for future studies on ASC
faculty salary.
The
study did not provide any evidence of gender differences in salary, even though
the literature indicates that women faculty earn less on average than their male
counterparts (Bellas, et. al. 2001; Hamton, et. al. 2000). Bellas, et al.
(2001), however, indicated in their study that this gap maybe partially
attributed to the concentration of women faculty in relatively low-paying
disciplines. It might be a reason for gender not being a predictor of salaries
for the faculty in construction schools.
References
Bellas,
M. L., Ritchey, P. N., & Permer, P. (2001). Gender differences in the
salaries and salary growth rate of university faculty: An exploratory study. Sociological
Perspectives, 44(2), 163.
Cox,
A. M. (2001). Law professors top list in study comparing salaries across
disciplines. Chronicle of Higher Education, 47(46), A10.
Ferber,
M. A. (1974). Professors, performance, and rewards. Industrial Relations,
13(1), 69-77.
Hampton,
M. & Oyster, C. (2000). Gender inequity in faculty pay. Compensation and
Benefits Review, 32(6), 54-59.
Information
Management & Testing Services (1996). http://www.baylor.edu/IMTS/Vol967/IR96714.htm:
Faculty Salary Comparisons 1991-92 through 1995-1996.
IRT Series, 96-97(14). Waco, TX: Baylor University.
Lamb,
S. W. & Moates, W. H. (1999). A model to address compression inequities in
faculty salaries. Public Personnel Management, 28(4), 689-700.
Monks,
J. & Robinson, M. (2001). The returns to seniority in academic labor market.
Journal of Labor Research, 22(2), 415-430.
Moore,
J. M., Newman, R. J., & Turnbull, G. K. (1998). Do academic salaries decline
with seniority? Journal of Labor Economics, 16(2), 352-364.
Ott,
R. L. (1993). An introduction to
statistical methods and data analysis, (4th ed.) Belmont, CA.
Wadsworth Publishing Company.
Schneider,
A. (1999). Law and finance professors are top earners in Academe survey finds. Chronicle
of Higher Education, 45(38), 14-19.
Webster,
A. L. (1995). Demographic factors affecting faculty salary. Educational and
Psychological Measurement, 55, 728-35.