Professional Documents
Culture Documents
Boudreau1997 Gender Rank Salary
Boudreau1997 Gender Rank Salary
3,1997
Many different approaches, almost ail of which use some form of regression, have
been used to study the issue of gender equity in university faculty salaries. One major
point of contention in ail of these approaches is whether faculty rank, which is univer-
sity conferred, should be included as a predictor variable. Two illustrations are pre-
sented to demonstrate how omitting faculty rank as a predictor variable from gender
equity studies of university faculty salaries can lead to incorrect conclusions concern-
ing gender discrimination. The first illustration uses hypothetical data constructed so
that there is no difference in salary due to gender. However, when faculty rank is not
included as a predictor variable in the regression model, there is a significant differ-
ence in salary due to gender. The second illustration uses actual data from a study of
gender equity in pay at Bowling Green State University. This data set is used to con-
struct a new data set that is totally free of gender bias. When a regression model omit-
ting faculty rank is fit to this gender bias-free data, again a significant difference in
salary due to gender is present. Therefore, it is recommended that faculty rank be in-
cluded as a predictor variable in any model used to study gender equity relating to
salary.
Nancy Boudreau, lames Sullivan, William Balzer, Ann Marie Ryan, Robert Yonker, Todd
Thorsteinson, and Peter Hutchinson, Bowling Green State University. Address correspondence to:
Nancy S, Boudreau, Department of Applied Statistics and Operations Research, Bowling Green State
University, Bowling Green, OH 43403.
297
036l-OM5/97rt>600-0297SI2.50m © 1997 Human Sciences Press, Inc.
298 BOUDREAUETAL
1993). The research literature is replete with debates around the conceptual or op-
erational definition of these predictor variables (e.g., the use of possession of the
terminal degree versus possession of the doctorate; Lassiter, 1983; Raymond,
Sesnowitz, and Williams, 1993; Riggs et al., 1986; Simpson and Rosenthal,
1982). By far the most controversial point in gender equity studies in higher edu-
cation, however, is whether to include faculty rank as a predictor of salary (Gray,
1985; Barrett and Sansonetti, 1988).
The primary goal of this paper is to demonstrate how the omission of faculty
rank as a predictor variable from a gender equity study of faculty salary can lead
to incorrect conclusions regarding gender discrimination. A brief overview of the
arguments for or against the inclusion of faculty rank in a salary model is pre-
sented followed by two demonstrations, the first using hypothetical salary data
and the second using actual salary data, supporting the inclusion of faculty rank
as a predictor of salary.
dividual candidates' credentials than would be expected at the entry assistant pro-
fessor level.
In some cases a candidate is hired prior to completion of all degree require-
ments (usually "all but dissertation"). In some of these cases the initial rank is
specified as instructor with a salary commensurately lower than what would be
paid to a candidate who has completed the terminal degree. Language in the letter
of appointment specifies a date when all degree requirements must be satisfied.
Completion of requirements causes the contract to convert automatically to the
assistant professor level with a specified salary increase; failure to complete de-
gree requirements by the specified date results in termination.
The other method through which rank is determined, and through which rank
affects salary, is the promotion process. Only probationary or tenured faculty are
eligible for consideration for promotion. The promotion process involves an in-
tensive review of the candidate's credentials and recommendations by the aca-
demic department, department chair, college review committee, dean, and vice
president for academic affairs. If promotion recommendations are accepted by the
president and the board of trustees, promotions in rank are accorded to faculty ac-
companied by permanent stipends added to their salaries, each rank having its
own specified salary increment.
promoted to full professor while the female has not. In addition, suppose the
yearly salary increases for the male have been higher than those for the female
(salary discrimination). Thus, the only differences in the set of predictor variables
for these two faculty members are rank and gender. The female is at a lower rank
than she should be, but her salary will be relatively high for that rank. Had she
been properly promoted, her salary would be lower than that of her male counter-
part because of the discrimination in yearly raises. If rank is included as a predic-
tor variable in the regression model, the magnitude of this difference in salary
will not appear as large as it really is. This could possibly lead to the conclusion
that no gender difference in salary exists when it really does.
In summary, excluding rank as a predictor of faculty salary is based on the ar-
gument that both rank and salary decisions by the university carry the same po-
tential for bias, and a salary model with rank included may underestimate salary
bias due to gender (Moore, 1993). For this reason, the "salary kit" method en-
dorsed by the American Association of University Professors recommends that
faculty rank should not be included in the model testing gender discrimination in
faculty pay (Gray, 1990; Gray and Scott, 1980; Scott, 1977).
eluded rank only after an analysis of whether or not rank was gender biased
(Ervin, Thomas, and Zey-Ferrell, 1984; Swartzman, Seligman, and McClelland,
1992; Riggs et al., 1986). For example, Riggs et al. (1986) and Swartzman et al.
(1992) used discriminant function analysis to test for gender biases in rank.
Another approach, used by Lassiter (1983), sorted faculty members by sex, pos-
session of terminal degree, years since terminal degree, and years of college ex-
perience, and conducted chi-square analyses to test for differences between male
and female faculty members. As might be expected, however, there is some de-
bate on what methods and criteria should be used to assess gender discrimination
in faculty rank:
The problem of what constitutes a proper study of whether rank is sex-linked embraces
not only the question of differential promotion rates but also that of whether initial ap-
pointments of similarly qualified men and women are made at different ranks. While no
definitive guide for such analysis exists, it is clear that simply looking at the outcome of
the decision on those actually considered for promotion is not enough. (Gray and Scott,
1980, p. 176)
A second argument for including faculty rank in faculty salary studies is that
omitting important predictor variables from salary studies may result in an under-
specified model, which makes it difficult to correctly interpret the salary model.
Specifically, if a meaningful predictor variable has been left oat of a salary
model, the effect of this predictor on the dependent variable is often shifted inap-
propriately to other predictors that have been included in the model, leading one
to underestimate the influence of the omitted predictor and overestimate the in-
fluence of those related and included predictor variables. Faculty rank has been
shown to be the single best predictor of salary (Osborne, 1990; Raymond, Ses-
nowitz, and Williams, 1988; Stacy, 1983).
Finally, Stacy and Holland (1984) argue that rank is often an important ex-
planatory variable for salary disparities between male and female faculty. There
are usually differences in the number of male and female faculty members at
higher ranks at universities, which may simply reflect the greater availability of
males in the applicant pool several decades ago. If one ignores these differences
in distribution of males and females across ranks, gender difference in salary (I.e.,
combining ail ranks, the average salary of female faculty was 82% of that for
males) was approximately double the actual difference observed when rank was
taken into account (i.e., female full, associate, and assistant professors averaged
90%, 95%, and 95% of the salaries of their male counterparts, respectively).
Moore's (1993) review of salary equity studies found that rank was included in
every case she examined. University-based case studies of salary discrimination
typically included rank (e.g,, Bergman and Scott, 1991; Damelson and Smith,
1981; Fisher, Motowidlo, and Werner, 1993; Raymond, Sesnowite, and Williams,
1988; Baker etal., 1988; Schau and Heyword, 1987).
FACULTY RANK IN STUDIES OF GENDER EQUITY 303
Summary
The weight of the evidence reviewed above suggests that it is both reasonable
and appropriate to include rank as a predictor to study gender equity in salary.
However, it would be prudent to support the inclusion of rank with an accom-
panying study to evaluate the presence or absence of discrimination in obtaining
faculty rank through promotion. Thus, in gender equity studies relating to salary
there are really two questions to be answered: "Are faculty members at the ap-
propriate ranks?" and "Is there a difference in salary due to gender?"
Rank
Assistant Professor Full Professor Marginal
Gender (code = 0) (code = 0) Means
Male $40K $32K $70k $70K
Faculty $64K $64K S55.2K
(code = 0) $56K $56k
$50K $50K
(cell mean $36K) (cell mean $60k)
Female $46K $46K $56K $64K
Faculty $40K $40K S40.8K
(code = 1) $32K S32K
$26K $26K
(cell mean $36K) (cell mean $60k)1
Marginal $36K $60k $60k
Means
male assistant professor salary and the average female full professor salary is the
same as the average male full professor salary. This "gender difference" occurs
because an important predictor variable related to salary, that is, rank, has been
omitted from die model. If rank and gender are both included in the model, the
parameter estimate for rank of $24,000 (i.e., the difference between the average
salaries for full and assistant professors) is statistically significant, t(17) = 5.62,
p = .000, and the parameter estimate for gender of $0 (Le., the difference between
the average salaries of female and male faculty members) is not statistically sig-
nificant, 1(17) = 0, p as l.o. Finally, if gender, rank, and the interaction between
gender and rank are included in the model, the parameter estimate for rank of
$24,000 is again significant while the parameter estimates for gender of $0 and
gender by rank of $0 are nonsignificant.
Now, suppose that systematic discrimination is introduced by reducing each fe-
male faculty salary $ 10K. Using only gender to predict salary, the parameter esti-
mate for gender is now -$24,400 (i.e., the difference between the average salary
of females collapsing across ranks after the $10K was subtracted and the average
salary of males collapsing across ranks) and is statistically significant, t(18) =
-4.35, p = .000. When both rank and gender are included in the model, the pa-
rameter estimate for rank of $24,000 (i.e., the difference between the average
salaries for full and assistant professors within each sex) is statistically signifi-
cant, *(17) = 5.62, p = .000, and the parameter estimate for gender of -$10,000
(i.e., the difference between the average salaries of female and male faculty mem-
bers within each rank) is statistically significant, t(17) = -2.34,p = .032. This pa-
FACULTY RANK IN STUDIES OF GENDER EQUITY 305
rameter estimate for gender reflects the systematic discrimination that was intro-
duced into the data.
In conclusion, the first simple hypothetical example demonstrates that by ex-
cluding from the salary model an important predictor variable such as faculty
rank that is highly related to salary, gender can appear to have a significant impact
on salary when, in fact, it does not. This leads one to draw an inappropriate con-
clusion about the presence of gender discrimination in salary. However, if sys-
tematic discrimination does exist, the inclusion of rank does not cover up the true
gender difference.
male salary model including gender is now fit to this new database, the parameter
estimate for gender would be close to zero (it would be exactly zero if no random
error term were added). Thus, is this modified database, the salary for each fac-
ulty member is equal to what he or she would be paid if everyone were paid ac-
cording to the male salary model.
The above procedure was used with the 1993-94 Faculty Salary Base at BGSU
to create a gender bias-free database that was then used to evaluate the necessity
of including rank in BGSU's salary model. This database contained a total of 511
male and 214 female faculty members. The proportions of male and female fac-
ulty members at the different faculty ranks were quite different (91% of the full
professors were males, 74% of the associate professors were males, 56% of the
assistant professors were males, and 36% of the instructors/lecturers were males).
A promotion study was also done to make sure faculty were at the appropriate
ranks.
Parameter Standard
Variable Estimate Error P VIF
INTERCEPT 40835.00 1782.49 0.000 0.00
PRIOREXP 671.77 175.13 0.001 12.57
YRBG -153.22 95.81 0.110 15.06
Rl 16875.00 1375.33 0.000 7.59
R2 7087.51 957.72 0.000 3.33
R4 -11691.00 1992.10 0.000 3.64
YRRANK 484.48 141.97 0.001 17.38
TERMDEG -2019.97 1661.13 0.225 3.23
SALFAC 15208.00 4264.17 0.000 6.89
GRADFAC 1965.59 845.24 0.021 2.11
ADMIN -2068.66 1845.05 0.263 1.92
EMINENT 27340.00 3908.10 0.000 1.48
FIRELAND 4459.36 2365.66 0.060 4.65
PRIOREXP*YRBG -84.10 17.42 0.000 10.60
PRIOREXP*R1 700.98 137.71 0.000 4.41
PRIOREXP*YRRANK 63.50 21.16 0.003 7.64
PRIOREXP*SALFAC -1523.31 391.21 0.000 1.19
YRBG*YRRANK -20.11 6.06 0.001 3.15
YRBG*SALFAC -1126.72 192.76 0.000 1.25
R1*YRRANK 768.26 141.04 0.000 6.98
R2*YRRANK 318.35 121.62 0.009 4.42
R4*TERMDEG 7408.07 3189.90 0.021 1.64
R4*SALFAC -19681.00 7905.44 0.013 1.76
YRRANK*ADMIN 883.05 201.34 0.000 1.51
TERMDEG*FIRELAND -5582.57 2731.43 0.042 4.66
SALFAC*GRADFAC 17868.00 4675.65 0.000 6.38
ADMIN*FIRELAND 26447.00 6183.44 0.000 1.24
F" 10U84,p = 0.000
AdjR2 = 0.836
W=511
s = 5547.16
5,547.16. The resulting random error terms are from a normal distribution with a
mean of zero and a standard deviation of 5,547.16.3
Parameter Standard
Variable Estimate Error P
INTERCEPT 36788.21 852.27 0.000
PRIOREXP 108.91 123.74 0.379
YRBG 819.65 49.32 0.000
TERMDEG 6447.79 1009.96 0.000
SALFAC 11901.08 3416.40 0.001
GRADFAC 7333.61 775.87 0.000
ADMIN 8141.68 1739.21 0.000
EMINENT 40945.44 4309.77 0.000
FIRELAND 446.31 1365.81 0.744
PRIOR*GRADFAC 646.20 142.03 0.000
YRBG*GRADFAC 148.87 59.36 0.012
SALFAC*GRADFAC 26436.73 3987.90 0.000
ADMIN*FIRELAND 23688.23 5470.57 0.000
GENDER -1642.86 636.70 0.010
F= 176.35, p = 0.000
AdjR2 = 0.759
N =725
s = 7055.75
mately 2.5% of the overall average faculty salary). The male salary model (that
includes rank and years in rank) including gender was then fit to this new "gen-
der-biased" database. The parameter estimate for gender is —$1,102 and is statis-
tically significant, f(697) = -2.128, p = .033 (it is not equal to -$1,200 because
the random error term was added). Thus, if gender bias does exist in the data set,
including rank and years in rank will not cover up this true difference. These ad-
ditional analyses, then, reaffirm that rank should be included in a salary model.
Although we used hierarchical multiple regression to demonstrate the need to
include faculty rank in models used to determine if gender bias exists in faculty
salaries, this is by no means the only way to analyze the data. Other methods such
as path analysis or structural equations could also be used.
CONCLUSION
If gender inequity exists in faculty salaries, it may be due to one or both of two
types of discrimination. First, female faculty may not be at the appropriate rank.
Second, even if female faculty are at the appropriate rank, a difference in salary
may exist. A promotion study needs to be done to determine if female faculty are
at the appropriate rank. Faculty rank needs to be included in a salary model to
even hope to uncover the second type of discrimination.
Two illustrations were presented to demonstrate how the omission of faculty
rank from faculty salary models could lead to the inappropriate conclusion that
gender inequity in salaries exists when it does not. In both examples, the propor-
tions of males and females at various faculty ranks are different. If the propor-
tions of males and females across ranks are similar, leaving rank out of the model
would probably not affect the significance or nonsigniflcance of gender. It could,
however, affect other model characteristics such as the model variance and the Rz
value of the model. Thus, it is recommended that faculty rank be included as a
predictor variable in any model used to study gender equity relating to salary. In
addition, it is also recommended that any gender equity study of salaries that in-
clude rank be accompanied by a promotion study. When bias in the awarding of
rank is found, such bias should be appropriately remedied.
The issues discussed in this paper may generalize to other situations where al-
legations of discrimination in salary are made. For example, a similar logic and
rationale could be applied concerning discrimination in faculty salary due to race.
If the proportions of minority and nonminority faculty members differ at the dif-
ferent faculty ranks, omitting rank from the model could lead to an inappropriate
conclusion regarding discrimination in salary due to race.
Because the issue of salary equity is so important, much care should be taken in
conducting salary equity studies. A thorough understanding of the effects of omit-
ting important predictor variables is needed by those who undertake such studies.
310 BOUDREAU ET AL.
NOTES
1. The following variables were used in the regression models:
PRIOREXP— Prior experience, found by subtracting the years at BGSU from the years since
highest degree
YRBG — Years at BGSU
YRRANK — Years in current rank
Rl — Rl = 1 if full professor, 0 if not
R2 — R2 = 1 if associate professor, 0 if not
R4 — R4 = 1 if instructor or lecturer, 0 if not
TERMDEG— TERMDEG = 1 if possess terminal degree in field, 0 if not
ADMIN — ADMIN = 1 if held a prior administrative appointment at BGSU other than de-
partment chair or was hired into university as department chair or school director,
0 if not
EMINENT — EMINENT = 1 if currently appointed as eminent scholar, 0 if not
GRA0FAC— GRADFAC= 1 if member of graduate faculty, 0 if not
FIRELAND— FIRELAND = 1 if appointment on Firelands campus, 0 if not
SALFAC — Measure of external market factor
GENDER — GENDER = 1 if female, 0 if not
2. The external market salary factor (SALFAC) was found using the 1993/94 national salary data pro-
vided by the National Association of State Universities and Land Grant Colleges (NASULGC).
Survey respondents included 77 of the 93 NASULGC institutions with at least five different doc-
toral programs. Because BGSU is a NASULGC institution, respondents in this survey represent a
reasonable set of peer institutions. SALFAC represents, for a given discipline/major field, the ra-
tio of the national average academic-year salary of full-time faculty for that discipline/major field
to the average academic year salary for full-time faculty of all disciplines/major fields. Since the
classification scheme used by NASULGC was too broad to accurately represent all specialty areas
within the College of Business Administration, salary data gathered in the same manner by the
American Assembly of Collegiate Schools of Business (AACSB) from university business
schools were used.
3. The male salary model plus gender was fit to this new "gender bias-free" database. The resulting
parameter estimate for gender, 97.82, is close to zero and is not statistically significant, ?{697) =
0.189, p = .850.
REFERENCES
Aiken, L. S., and West, S. G. (1991). Multiple Regression: Testing and Interpreting
Interactions. Newbury Park, CA: Sage.
Baker, D. E., Lattin, D. L., Siddons, L. M., and Thomas, J. (1988). Report of the faculty
services subcommittee: Study of male/female faculty salary differences. American
Journal of Pharmaceutical Education 52:421-424.
Balzer, W., Boudreau, N. S., Ryan, A. M., Snavely, D., Sullivan, J., and Yonker, R. (1995).
Gender Equity in Faculty Salary at Bowling Green State University: A Report Based on
1993-1994 Faculty Salary Data. Unpublished document, Office of the President,
Bowling Green State University, Bowling Green, OH.
FACULTY RANK IN STUDIES OF GENDER EQUITY 311
Barrett, G. V., and Sansonetti, D. M. (1988). Issues concerning the use of regression analy-
sis in salary discrimination cases. Personnel Psychology 41:503-515.
Bereman, N. A., and Scott, J. A. (1991). Using the compa-ratio to detect gender Mas in
faculty salaries. Journal of Higher Education 62: 556-569.
Chasin, S. H., Goldfarb, M. G., and LaNoue, G. R. (1989). An exploration of gender and
other influences on the earnings of pharmacy school faculty. American Journal of
Pharmaceutical Education 53: 1-6.
Cohen, J., and Cohen, P. (1983). Applied Multiple Regression/Correlation Analysis for the
Behavioral Sciences. Hillsdale, NJ: Lawrence Erlbaum.
Danielson, J. L., and Smith, R. (1981). The application of regression analysis to equality
and merit in personnel decisions. Public Personnel Management Journal 10: 126-131.
Ervin, D., Thomas, B. J., and Zey-Ferrell, M. (1984). Sex discrimination and rewards in &
public comprehensive university. Human Relations 37: 1005-1025.
Ferber, M. A., and Green, C. A. (1982). Traditional or reverse sex discrimination? A case
study of a large public university. Industrial and Labor Relations Review 35:550-564.
Fisher, B. D., Motowidlo, S., and Werner, S. (1993). Effects of gender and other factors on
rank of law professors in colleges of business: Evidence of a glass ceiling. Journal of
Business Ethics 12: 771-778.
Fogel, W. (1986). Class pay discrimination and multiple regression proofs. Nebraska Law
Review 65: 289-329.
Gray, M. W. (1985). Legal perspectives on sex equity in faculty employment. Journal of
Social Issues 41:121-134.
Gray, M. W. (1990). Achieving pay equity on campus. In E. L. Scott (ed.), Higher
Education Salary Evaluation Kit. Washington, DC: American Association of University
Professors.
Gray, M. W., and Scott, E. L. (1980). A "statistical" remedy for statistically identified dis-
crimination. Academe 66: 174-181.
Johnson, C. B., Riggs, M. L., and Downey, R. G. (1987). Fun with numbers: Alternative
models for predicting salary levels. Research in Higher Education 27:349-362.
Lassiter, R. L., Jr. (1983). The development and use of a faculty salary model for a higher
education system. Research in Higher Education 18: 333-358.
Moore, N. (1993). Faculty salary equity: Issues in regression model selection. Research in
Higher Education 34:107-126.
Neter, J., Wasserman, W., and Kutner, M. H. (1989). Applied Linear Regression Models,
(2nd ed.). Homewood, IL: Irwin.
Oaxaca, R. (1973) Male-female wage differentials in work labor markets. International
Economic Review 14: 693-709.
Osborne, J. S. (1990). A Strategy Towards Accountability: Identifying Faculty Salary
Policies at Kentucky Public Regional Institutions. Unpublished doctoral dissertation,
University of Kentucky.
Raymond, R. D., Sesnowitz, M. L., and Williams, D. R. (1988). Does sex still matter? New
evidence from the 1980s. Economic Inquiry 26:43-58.
Raymond, R. D., Sesnowitz, M. L., and Williams, D. R. (1993). Further evidence on gen-
der and academic rank. Quarterly Review of Economics and Finance 33:197-215.
312 BOUDREAU ET AL.