17
The Social Science Journal 42 (2005) 165–181 Gender differences in faculty pay and faculty salary compression Kathleen Burke a , Kevin Duncan b,, Lisi Krall a , Deborah Spencer a a SUNY, Cortland, USA b Colorado State University-Pueblo, School of Business, 2200 Bonforte Blvd., Pueblo, CO 81001-4901, USA Abstract A recording distinction between cost-of-living and merit adjustments at a unionized, public liberal arts college allows us to examine several issues related to gender differences in faculty pay. For example, we find that annual fixed-dollar merit increases and similar starting salaries contribute to comparable salary growth rates for female and male faculty. In this setting, the male faculty earnings advantage is traced to higher rank and years of service. These results underscore the importance of gender-neutral salary- setting practices and equal access to promotion and retention for female faculty. The salary distinctions also allow us to determine the source of the seniority penalty. The economics literature is divided on whether the often-observed lower pay of senior faculty is deserved. We find that merit pay rises with additional years of seniority and that the seniority penalty is rooted in cost-of-living adjustments that fail to keep pace with market trends. These findings illustrate how the seniority penalty can be linked to budget considerations rather than the lower productivity of senior staff. © 2005 Elsevier Inc. All rights reserved. 1. Introduction Gender differences in pay and salary compression at academic institutions are issues that are often addressed separately in the literature. This paper examines both by taking advantage of an administrative practice at a unionized, public, liberal arts college (hereinafter, U-PLAC) of separating annual cost-of-living increases from merit awards. These data allow us to examine gender differences in total faculty salary and its components; cost-of-living and promotion adjusted starting salary as well as accumulated merit pay. These data also allow us to examine the effects of cost-of-living adjustments and merit awards on the compression of total faculty salary. Corresponding author. Tel.: +1 719 549 2228. E-mail address: [email protected] (K. Duncan). 0362-3319/$ – see front matter © 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.soscij.2005.03.006

Gender differences in faculty pay and faculty salary compression

Embed Size (px)

Citation preview

Page 1: Gender differences in faculty pay and faculty salary compression

The Social Science Journal 42 (2005) 165–181

Gender differences in faculty pay and facultysalary compression

Kathleen Burkea, Kevin Duncanb,∗, Lisi Krall a, Deborah Spencera

a SUNY, Cortland, USAb Colorado State University-Pueblo, School of Business, 2200 Bonforte Blvd., Pueblo, CO 81001-4901, USA

Abstract

A recording distinction between cost-of-living and merit adjustments at a unionized, public liberal artscollege allows us to examine several issues related to gender differences in faculty pay. For example, wefind that annual fixed-dollar merit increases and similar starting salaries contribute to comparable salarygrowth rates for female and male faculty. In this setting, the male faculty earnings advantage is tracedto higher rank and years of service. These results underscore the importance of gender-neutral salary-setting practices and equal access to promotion and retention for female faculty. The salary distinctionsalso allow us to determine the source of the seniority penalty. The economics literature is divided onwhether the often-observed lower pay of senior faculty is deserved. We find that merit pay rises withadditional years of seniority and that the seniority penalty is rooted in cost-of-living adjustments thatfail to keep pace with market trends. These findings illustrate how the seniority penalty can be linked tobudget considerations rather than the lower productivity of senior staff.© 2005 Elsevier Inc. All rights reserved.

1. Introduction

Gender differences in pay and salary compression at academic institutions are issues that areoften addressed separately in the literature. This paper examines both by taking advantage ofan administrative practice at a unionized, public, liberal arts college (hereinafter, U-PLAC) ofseparating annual cost-of-living increases from merit awards. These data allow us to examinegender differences in total faculty salary and its components; cost-of-living and promotionadjusted starting salary as well as accumulated merit pay. These data also allow us to examinethe effects of cost-of-living adjustments and merit awards on the compression of total facultysalary.

∗ Corresponding author. Tel.: +1 719 549 2228.E-mail address: [email protected] (K. Duncan).

0362-3319/$ – see front matter © 2005 Elsevier Inc. All rights reserved.doi:10.1016/j.soscij.2005.03.006

Page 2: Gender differences in faculty pay and faculty salary compression

166 K. Burke et al. / The Social Science Journal 42 (2005) 165–181

Numerous studies have identified the factors contributing to differences in total salariesbetween female and male faculty employed at colleges and universities in the U.S. For example,Barbezat (1987, 1989)andToutkoushian (1998)suggest that from 90 to 93% of the genderdifference in total faculty salaries can be explained by gender differences in characteristicssuch as experience, productivity, rank, geographic location and personal as well as universityattributes.1 While these studies provide an estimate of salary discrimination in academia (animplied 7–10%), they offer limited insight into how specific university practices contribute tolower female salaries.

Examinations of the components of total salary and of campus salary-setting practiceshave been useful in gaining insight into this process. For example,Bellas, Ritchey, Neal, andParmer (2001)focus on gender differences in faculty salary growth that consists of annualcost-of-living and merit adjustments. These authors attribute the higher salary growth rate ofwomen employed at their campus, in part, to the salary-setting practice of awarding fixed-dollarannual increases. Such salary increases have a disproportionate impact on the earnings of thelowest paid faculty, many of whom are women. These authors also note that the practice ofindividualized bargaining with respect to starting salaries may contribute to earnings disparity atthe inception of careers. Men may be able to negotiate for higher starting salaries for a variety ofreasons including the segregation of women into low growth fields of study, gender differencesin productivity, or because male breadwinner ideologies result in greater acceptance of men’snegotiations. Regardless of the source of gender differences in starting salaries,Reskin, Liddy,Haignere and Frances (1992)argue that institutional salary-setting practices, such as percentageannual increases, perpetuate an initial male salary advantage and widen the absolute genderdisparity over time.Reskin et al. (1992)also note that allowing salaries to compress betweenmore and less senior faculty may have a disproportionate impact on females, if they are lessmobile.

The existence of compressed, or inverted, salaries at many campuses in the U.S. is wellknown. However, debate exists over the cause of this salary pattern. For example,Ransom(1993)argues that more senior faculty receive lower pay because of the monopsony powerenjoyed by universities.Bok (1993) and Bereman and Lengnick-Hall (1994)add thatdemand/supply conditions for new positions, coupled with limited budgets, contribute to therelatively lower pay of senior staff. These studies share the implication that the senioritypenalty is undeserved. However,Moore, Newman, and Turnbull (1998)find that their estimateof the seniority penalty disappears with controls for detailed measures of faculty produc-tivity. This finding implies that the penalty is due to the lower productivity of more seniorfaculty.

Salary-setting practices at U-PLAC include fixed-dollar merit increases along with admin-istrative policies that discourage the kind of individualized bargaining that contributes to aninitial male advantage. Given this setting, we measure gender disparity in total salary, cost-of-living and promotion adjusted starting salary (hereinafter, COLA salary) and accumulatedmerit pay. Furthermore, we use the method developed byOaxaca (1973)to decompose thegap in each salary component to illustrate how differences in faculty characteristics contributeto earnings disparity, once gender biases in salary-setting practices have been eliminated. Wealso estimate the relation between faculty seniority and each salary component to gain insightinto a source of salary compression.

Page 3: Gender differences in faculty pay and faculty salary compression

K. Burke et al. / The Social Science Journal 42 (2005) 165–181 167

The remainder of this paper is organized as follows. The U-PLAC data and our faculty salaryregression models are discussed in the following section. Our estimated results are presentedand discussed in subsequent sections. We conclude with a discussion of the applicability ofour methods and conclusions to other campuses.

2. Data and models

The U-PLAC data include tenured and tenure-track faculty who have completed at leastone year of service at this institution. These data contain information on faculty salaries(for the 1998–1999 academic year), merit increases, performance, and years of serviceat the present institution and years of prior experience. The Office of Human Resourcesat U-PLAC collected this information. We do not include new hires because these fac-ulty have not participated in merit evaluations at U-PLAC. Furthermore, the sample doesnot include individuals who hold administrative positions above the level of departmentchair.

As mentioned previously, these data contain a record of accumulated merit pay from theinception of each faculty member’s career at this institution. Because of this recording distinc-tion, we are able to examine the gender wage gap among these faculty from three perspectives.That is, gender disparities that arise from differences in total salaries, COLA salaries, andaccumulated merit pay.2

Salary-setting practices at U-PLAC do not have a significant negative impact on the earningsof female faculty. As mentioned above, annual merit increases are a fixed-dollar amount.Consequently, equal merit awards to female and male faculty preserve disparity in absoluteterms. However, such awards reduce disparity when measured as a percent.3 Cost-of-livingadjustments on this campus are based on a percent of faculty base salary. This type of annualadjustment may add to an initial male salary advantage, but, starting salary negotiations atthis campus do not offer the kind of flexibility that results in significantly higher salaries fornew male faculty.4 Evidence of this is revealed by gender differences in starting salaries forrecently hired U-PLAC faculty. For example, in 1998, new male faculty received an averagestarting salary of $36,680 while new female faculty received an average starting salary of$36,622. The $58 difference in these starting salaries is not statistically significant. Given thisrough equity at the inception of careers, percentage cost-of-living increases do not have asignificant negative impact on the relative earnings of female faculty as careers develop at thisinstitution.

To illustrate the advantages of these data, we estimate the following three faculty earningsequations separately for female and male faculty:

• Model 1

TOTAL SALARY = β0 + β1 SENIORITY+ β2 SENIORITY2

+β3 PRIOR EXPERIENCE+ β4 X + β5 CUPA SAL

+β6 TOP PERFORM+ µ

Page 4: Gender differences in faculty pay and faculty salary compression

168 K. Burke et al. / The Social Science Journal 42 (2005) 165–181

• Model 2

COLA SALARY = β0 + β1 SENIORITY+ β2 SENIORITY2

+β3 PRIOR EXPERIENCE+ β4 X + β5 CUPA SAL

+β6 TOP PERFORM+ µ

• Model 3

ACCUMULATED MERIT = β0 + β1 SENIORITY+ β2 SENIORITY2 + β3 X

+β4 CUPA SAL+ β5 TOP PERFORM+ µ

where the dependent variable for Model 1, TOTAL SALARY, is the faculty member’s con-tractual salary inclusive of accumulated merit, but minus stipends for department chairs. ForModel 2, the dependent variable (COLA SALARY) is contractual salary minus stipends, mi-nus past merit increases. This variable is a measure of the faculty member’s promotion andinflation adjusted starting salary. For Model 3 the dependent variable is the faculty member’saccumulated merit pay (ACCUMULATED MERIT). Merit pay at U-PLAC is awarded annu-ally based on a faculty member’s performance evaluation that is weighted between teaching,scholarship and service. The record of annual merit awards, from the inception of each facultymember’s U-PLAC career, is used to derive our measure of accumulated merit pay. The def-initions, means and standard deviations for all variables used in the estimation of these threemodels are reported inTable 1by gender. The implications of these statistics are discussed inSection3 below.

The usual approach in the literature is to estimate the natural log of salary as a function offaculty characteristics. However, since accumulated merit pay for some of the faculty in oursample is $0, the semi-log approach is not suitable for Model 3. Therefore, to facilitate thecomparison of coefficients across models, we estimate untransformed faculty salaries for allthree models. These untransformed results do not differ significantly from those obtained fromthe semi-log specification.5

The specification of the right-hand side of the equations is typical of other studies that haveexamined faculty salaries. For example,Ransom (1993)andBrown and Woodbury (1998)use the quadratic form of the years of seniority (years of service at an institution) indicatedby SENIORITY and SENIORITY2. The signs on the coefficients of the seniority variablesindicate faculty salaries that rise, or fall, with additional years of service at an institution. Forexample, a negative sign for the linear term and a positive coefficient for the quadratic termindicate salary inversion and compression at this campus. PRIOR EXPERIENCE is the numberof years of work experience a faculty member accumulated prior to coming to U-PLAC.X isa vector of faculty characteristics including current rank, race, position as chair, and degreestatus.

The data for CUPA SAL were gathered from the 1998 College and University PersonnelAdministration (CUPA) market salary for new assistant professors for each faculty disciplineat U-PLAC. As an independent variable in the estimated equation, CUPA SAL measuresthe relationship between U-PLAC faculty salaries and external (market), entry-level salaries.Brown and Woodbury (1998)use entry-level, field-specific market salaries to determine if

Page 5: Gender differences in faculty pay and faculty salary compression

K. Burke et al. / The Social Science Journal 42 (2005) 165–181 169

Table 1Variable descriptions and summary statistics for female and male faculty

Variable Description Female Male

TOTAL SALARY 1998–1999 salary fortenured, tenure-trackfaculty (minusstipends)

$43,974.07a (8482.93) $51,736.20 (10524.87)

COLA STARTING SALARY 1998–1999 total salaryminus accumulatedmerit (minus stipends)

$40,548.63a (6005.64) $46,129.25 (9021.84)

ACCUMULATED MERIT Accumulated merit fortenured, tenure-trackU-PLAC faculty

$3425.45a (3482.48) $5606.95 (3413.52)

SENIORITY Full-time years ofservice at U-PLAC

12.50a (8.97) 21.15 (9.72)

PRIOR EXPERIENCE Full-time years ofservice prior toU-PLAC

0.14 (0.77) 0.13 (0.73)

RANK Dummy variables fordistinguished, full andassociate professorswith assistants as thereference category

0.04 (0.50), 0.23a

(0.43), 0.46 (0.19)0.07 (0.25), 0.49 (0.49),0.38 (0.50)

PHD Equal to one for Ph.D.sand zero for M.A.

0.91 (0.29) 0.91 (0.29)

CHAIR Equal to one fordepartment chairs, 0otherwise

0.04a (0.19) 0.14 (0.34)

CUPA SAL Average CUPA salaryfor new assistants bydiscipline

$37,499.54 (2609.67) $37,648.00 (2497.26)

RACE One if non-white, zeroif white

0.04a (0.19) 0.16 (0.37)

TOP PERFORM Equal to one for topperforming faculty for1998–1999, 0otherwise

0.41 (0.50) 0.38 (0.49)

N 56 133

Source: Salary data for U-PLAC tenured and tenure-track faculty. Standard deviations are given in parentheses.a The mean for females is significantly different at the .05 level than the comparable mean for males.

changes in external salaries are transmitted to faculty salaries at an institution. These authorsfind that if the market salary of a new male economist increases by 1%, the salaries of economicsprofessors at their institution increase by 0.65%. We include CUPA SAL in our estimates ofmale and female faculty salaries to determine if the relationship between external and internalsalaries at U-PLAC differs by gender. Since CUPA SAL matches with any faculty member’sfield, we omit department variables to avoid redundancy in the results reported below. However,we also estimated the models with other controls for departments for comparison purposes.The results of these estimates are discussed below.6

We include two measures of faculty performance in an attempt to control for the effect ofproductivity on salary. First, TOP PERFORM is equal to one if the faculty member was among

Page 6: Gender differences in faculty pay and faculty salary compression

170 K. Burke et al. / The Social Science Journal 42 (2005) 165–181

the top performers at this institution, based on the previous year’s faculty evaluations. TOPPERFORM is zero for those who were not included in this group. The published evaluationguidelines issued by the Office of the President at U-PLAC indicate that productivity at U-PLACis based on the weighted performance in teaching, research and service. Exemplary teachingand service are recognized, however, research, in terms of the number of publications, carries ahigher weight in performance decisions. Thus, TOP PERFORM is a broad measure of facultyperformance. TOP PERFORM not only captures an individual’s productivity in terms of theirperformance within the institution (through teaching and service evaluation), but this variableis also an indication of mobility and marketability (indicated by publications). The limitation ofthis measure of productivity is that we use data for only one year’s performance to select TOPPERFORM. However, it is very likely that faculty recognized as top performers in the selectedyear have a history of high productivity. For example, faculty recognized as top performersmay have more accumulated merit pay, or may have received higher starting salaries becauseof a record of high performance. This example would be supported by a coefficient for TOPPERFORM that exceeds the average annual performance increase in pay, suggesting that topperformers have an earnings (and productivity) record extending beyond a specific annualaward.7

Among the variables in the vectorX is CHAIR, a measure for those holding a positionas department chair (equals one if a chair, else 0). If more productive faculty move into thesepositions, this variable will also measure performance. Results of previous research concerningthe earnings effect of holding a position as chair are mixed. For example,Katz (1978)does notfind a significant chair effect, however,Siegfried and White (1973)report a positive impact onsalary. The error term in all models isµ.

We apply the wage decomposition technique developed byOaxaca (1973)to the modelestimates to gain further insight into the causes of gender disparity in each of the salarycategories.8 Oaxaca’s method decomposes a wage gap into “explained” and “unexplained”portions. The explained portion is due to gender differences in faculty attributes measuredby gender differences in the average values of the variables used in the salary estimates. Theunexplained portion of a wage gap is due to gender differences in how faculty attributes arerewarded. This portion is due, primarily, to gender differences in the estimated coefficients,thus providing evidence of different reward systems for men and women. Hence, it is commonto attribute this unexplained portion to discrimination.9

Gender differences in rank among U-PLAC faculty may be due to lower female facultyproductivity, or to discriminatory promotion practices. Consequently, caution should beexercised when interpreting the explained portion of the gap that is due to differencesin rank. Barbezat (1991)argues that if women face discrimination in promotion, it isinappropriate to include measures of rank when using the decomposition method. Weinclude measures of rank in our estimates to determine how much of the gap can beexplained by differences in these variables, regardless of the cause of gender differencesin their values. We also discuss the results of an estimate that does not include measuresof rank. Additionally, we conduct Chow tests to determine if structural similarities ex-ist between female and male estimates. We test for specific differences in coefficientsby utilizing female dummy interaction terms. The results of these tests are discussedbelow.

Page 7: Gender differences in faculty pay and faculty salary compression

K. Burke et al. / The Social Science Journal 42 (2005) 165–181 171

3. Results

The variables included in our models that measure gender differences in individual andjob-related characteristics that contribute to gender differences in salaries, are reported inTable 1. The salary data indicate that in every category, female faculty at U-PLAC earn lessthan male faculty. These differences are statistically significant at the .05 level. These dataindicate that female faculty earn 85% of male total salary, 87.9% of COLA salary and 61.1%of the accumulated merit pay of male faculty.

Gender differences in the latter two salary categories are particularly sensitive to genderdifferences in years of seniority and rank at this institution. For example, male faculty havesignificantly more years of seniority, are less likely to be assistant professors, but more likelyto be full professors.10 As a consequence, COLA salaries for men will be higher due to a longerrecord of cost-of-living adjustments and because of the salary steps associated with promotion.Accumulated merit pay may also be higher for males due to a longer record of meritoriousservice at this institution.

Data reported inTable 1indicate the absence of statistically significant differences withrespect to prior experience and the percentages of the females and males that hold terminaldegrees and positions as associate and distinguished professors. Men are significantly morelikely to be non-white and to hold positions as department chairs. The average CUPA salary bydiscipline indicates a narrow and statistically insignificant external starting salary gap, giventhe distribution of fields at U-PLAC. Finally, about 40% of the female and male samples aretop performers based on the previous year’s productivity.

3.1. Total salary results

Regression results for Model 1 are reported inTable 2. Results of a Chow test, and of anestimate with the interaction of a dummy variable for gender with each of the independentvariables, indicate that the female and male equations are structurally similar and that noneof the regression coefficients differ significantly by gender.11 Female and male coefficientsfor seniority and its square imply salary inversion and compression at this institution. How-ever, thet-values for the linear seniority terms fail to achieve conventional levels of statisticalsignificance. U-PLAC faculty with prior experience do not earn significantly more. Full anddistinguished professors earn significantly more (at the .01 level) than the reference category(assistant professors). In spite of a promotion salary step, associate professors do not earn morethan assistants, providing further evidence of salary compression at U-PLAC. Female facultywith terminal degrees earn significantly more than other women, while department chairs andnon-whites, of either gender, do not receive significantly different salaries. The salaries offemale and male faculty are not significantly related to the external market (as indicated by thet-values for CUPA SAL). The absence of a significant relation between external and internalsalaries also contributes to salary compression on this campus.

Top performing women earn more than lower performing female faculty. The differenceis significant at the .01 level. This result may be due to the combined effects of fixed-dollarmerit awards and the starting salaries that high performers receive within the female sample.Bellas et al. (2001)have illustrated the relative impact of fixed-dollar adjustments between

Page 8: Gender differences in faculty pay and faculty salary compression

172 K. Burke et al. / The Social Science Journal 42 (2005) 165–181

Table 2Total salary estimates for female and male, tenured and tenure-track U-PLAC faculty and percent of gender gapexplained by component

Variable Model 1: dependent variable = total salary− stipends

Female coefficient Male coefficient Percent of gender gap explaineda

Constant 18376.58 26873.91∗ –SENIORITY −345.40 −18.32 −2.04SENIORITY2 21.38** 13.38** 52.71PRIOR EXPERIENCE −314.32 394.95 −0.05ASCT. PROF. 1589.96 3630.52 −3.74FULL PROF. 10672.58∗ 11203.45∗ 37.53DIST. PROF. 16149.91∗ 20730.05∗ 8.01PHD 4654.30** 2045.20 0.00CHAIR 243.14 −1064.44 −1.37RACE −1350.49 −908.47 −1.40CUPA SAL 0.41 0.20 0.38TOP PERFORM 3855.24∗ 1474.28 −0.57

Total gender gap explained 89.46b

N 56 133R2 (adj) 0.754 0.705F 16.34 29.71

a Percent of the gender gap explained (by component) =βm(Xm −Xf )/(Wm −Wf ), whereβm = the regressioncoefficient for males, (Xm −Xf ) = the difference between male and female variable averages and (Wm −Wf ) is thedifference between male and female earnings for this salary measure ($7,762.13).

b Total gender gap explained is the sum of components, or the total percent of the gender gap explained bydifferences in the male and female faculty attributes.

∗ Statistically significant at the .01 level.∗∗ Statistically significant at the .05 level.

female and male faculty. Similarly, we would expect fixed-dollar awards to have an even largerdisproportionate impact among the lowest paid (female) faculty. In addition, high performingfemale faculty may be recruited at higher starting salaries relative to other female faculty. Thehigher overall salaries within the male sample likely explain why we do not observe a similarimpact for top performing male faculty.

Results from the Oaxaca wage decomposition for this model indicate that 89.5% of theearnings difference in total salaries between female and male faculty can be explained bydifferences in the mean values of the variables included in the estimated equation. This impliesthat 10.5% of the gap in total salary can be attributed to discrimination. This estimate iscomparable to the results reported byBarbezat (1987, 1989)andToutkoushian (1998)whoattribute from 7 to 10% of the gender difference in total faculty salaries to discrimination.

We also report inTable 2the percent of the gap explained by each variable. This allows usto examine the effect of each individual variable, or the combined effect of selected variables.For example, differences in years of seniority and rank have the largest effect and account for92.5% of the gap in total salary.12 The gender differences in years of seniority and rank maybe due to gender differences in productivity, the temporal dynamic of the entry of women intoacademia, or to discriminatory retention and promotion practices at U-PLAC. The results forTOP PERFORM, however, indicate that−0.6% of the gap is due to gender differences in recent

Page 9: Gender differences in faculty pay and faculty salary compression

K. Burke et al. / The Social Science Journal 42 (2005) 165–181 173

performance. This portion of the explained differential is negative because the female meanfor TOP PERFORM (0.41) is greater than the mean for male faculty (0.38). Taken together,the results for performance, seniority and rank suggest that this institution is aggressivelyrewarding the performance of females and that the gender differences in rank and seniority aredue to past discrimination, the more recent entry of women into this workplace and/or lowerfemale productivity in previous years. If this is the case, the effects of rank and seniority shoulddiminish as the current, highly productive female faculty advance in rank at this institution.Toutkoushian (1998)mirrors this optimism for recent female faculty based on results from anational survey of faculty. Regardless, the results of the decomposition analysis suggest thatalmost all of the gap in total salary would disappear with a more equitable distribution of rankand years of seniority at U-PLAC.

As discussed above,Barbezat (1991)argues that if discrimination affects the promotion offemale faculty, measures of rank should be omitted from the estimate when using the Oaxacadecomposition method. When the rank variables are omitted from our estimation we find thatthe explained portion of the gender gap falls by approximately six percentage points to 83.1.13

3.2. COLA salary results

Regression results for Model 2 are reported inTable 3. The dependent variable in thisspecification is COLA salary (total salary minus stipends, minus accumulated merit pay). Thecoefficients andt-values for the seniority variables indicate the presence of significant inversionand compression at U-PLAC with respect to this salary component. The seniority coefficientsare significant at the .05 level and their signs suggest a U-shaped COLA salary-seniority profile.The minimum of this profile occurs at 12.6 years of seniority for women and 12.2 years formen.

These results support the view that the seniority penalty is linked to supply/demand condi-tions for new hires, budget constraints, or to the monopsony power exploited by universities.For example, the downward sloping portion of the U-shaped profile suggests inverted salarieswhere more recently hired faculty receive offers at, or close to, prevailing market rates, whilethe cost-of-living adjustments for more senior faculty have failed to keep up with market trends.The upward sloping portion of the U-shaped profile suggests compressed salaries between themost senior staff and junior faculty. We observe these trends even with controls for facultyproductivity. This trend in COLA salaries serves as a strong incentive for senior faculty toseek employment elsewhere. However,Ransom (1993)argues that faculty mobility costs andmonopsony power enforce a university’s ability to penalize seniority without faculty turnover.

The regression intercept and the coefficients for the seniority variables suggest a lower andsteeper COLA salary profile for female faculty. However, results of a dummy interaction testindicate that the constant and coefficients from the two equations do not differ significantly.The results of a Chow test also indicate structurally similar estimates. These results suggestsimilarly sloped and positioned wage-tenure profiles for this salary component for female andmale faculty. This finding is consistent with the impact of campus policies that discouragethe kind of individualized bargaining that results in higher starting salaries for men. We wouldexpect similar profiles with respect to this salary category for female and male staff that receivesimilar starting salaries and annual COLA adjustments.

Page 10: Gender differences in faculty pay and faculty salary compression

174 K. Burke et al. / The Social Science Journal 42 (2005) 165–181

The results with respect to prior experience, rank, degree status, chair, race and top perform-ers are consistent with the results reported for total salary inTable 2. CUPA SAL is significantfor females possibly indicating that the influx of women into academia causes their COLAsalary to be more closely tied to the market salary.

The results of the decomposition analysis indicate that 85.6% of the gender gap in COLAsalaries can be explained by differences in the variable averages. That is, of the $5,580.62gender difference in this salary category, $4,777.01 can be explained by gender differencesin average characteristics. Consequently, $803.62 is unexplained. This finding is consistentwith the results of a practice of establishing roughly equal starting salaries for female andmale faculty. Almost all of the gender difference in this salary category can be explained bydifferences in faculty characteristics while very little of the gap is unexplained, or can beattributed to discriminatory preferences for men in the determination of starting salaries.

3.3. Accumulated merit results

While the results fromTable 3indicate a seniority penalty with respect to COLA salaries,the quadratic specification of years of seniority reported inTable 4indicates that accumulatedmerit pay increases with seniority at a diminishing rate.14 If merit pay at this campus is awarded

Table 3COLA salary estimates for female and male, tenured and tenure-track U-PLAC faculty and percent of gender gapexplained by component

Variable Model 2: dependent variable = COLA salary = total salary− stipends− merit

Female coefficient Male coefficient Percent of gender gap explaineda

Constant 18346.02** 32947.46∗ –SENIORITY −798.20** −631.04** −97.81SENIORITY2 31.63∗ 25.93∗ 142.09PRIOR EXPERIENCE −110.30 869.85 −0.16ASCT. PROF. 1595.60 4022.40 −5.77FULL PROF. 7827.39∗ 9116.30∗ 42.47DIST. PROF. 8536.31** 16575.99∗ 8.91PHD 3810.35 739.28 0.00CHAIR −2062.97 −2094.36 −3.75RACE −1700.10 −419.54 −0.90CUPA SAL 0.47** 0.13 0.35TOP PERFORM 2079.52 −259.82 0.14

Total gender gap explained 85.57b

N 56 133R2 (adj) .623 .588F 9.28 18.09

a Percent of the gender gap explained (by component) =βm(Xm −Xf )/(Wm −Wf ), whereβm = the regressioncoefficient for males, (Xm −Xf ) = the difference between male and female variable averages and (Wm −Wf ) is thedifference between male and female earnings for this salary measure ($5,580.62).

b Total gender gap explained is the sum of components, or the total percent of the gender gap explained bydifferences in the male and female faculty attributes.

∗ Statistically significant at the .01 level.∗∗ Statistically significant at the .05 level.

Page 11: Gender differences in faculty pay and faculty salary compression

K. Burke et al. / The Social Science Journal 42 (2005) 165–181 175

Table 4Accumulated merit estimates for female and male, tenured and tenure-track U-PLAC faculty and percent of gendergap explained by component

Variable Model 3: dependent variable = accumulated merit pay

Female coefficient Male coefficient Percent of gender gap explaineda

Constant −222.83 −5437.23** –SENIORITY 468.49∗ 651.36∗ 258.27SENIORITY2 −10.51** −13.28∗ −186.16ASCT. PROF. −99.14 −532.70 1.95FULL PROF. 2680.64∗ 1984.08** 23.65DIST. PROF. 7463.23∗ 3984.28∗ 5.48PHD 794.87 1344.06** 0.00CHAIR 2230.66 680.73 3.12RACE 349.98 −436.26 −2.40CUPA SAL −0.57 0.04 0.27TOP PERFORM 1803.99∗ 1795.35∗ −2.47

Total gender gap explained 101.72b

N 56 133R2 (adj) .755 .659F 17.94 26.52

a Percent of the gender gap explained (by component) =βm(Xm −Xf )/(Wm −Wf ), whereβm = the regressioncoefficient for males, (Xm −Xf ) = the difference between male and female variable averages and (Wm −Wf ) is thedifference between male and female earnings for this salary measure ($2,181.50).

b Total gender gap explained is the sum of components, or the total percent of the gender gap explained bydifferences in the male and female faculty attributes.

∗ Statistically significant at the .01 level.∗∗ Statistically significant at the .05 level.

on the basis of faculty performance, the concave merit–seniority profile is consistent with thehuman capital prediction that wages grow more quickly for younger workers who are engagedin the training that increases productivity and pay (seeBecker, 1975). This concave profilealso suggests that the accumulated merit of the most senior faculty is lower than the total meritpay of faculty in earlier stages of their careers. This could be due to the lower productivity offaculty in the advanced stages of their careers.1

The merit–seniority profile maximums suggest that accumulated merit pay rises until 22.3years seniority for females and 24.5 years for males. The maximum values of years of seniorityare 33 for women and 43 years for men. The years associated with rising merit pay overlap withthe years characterized by the seniority penalty with respect to the COLA salaries reported inTable 3. COLA salaries fall until approximately 12 years of seniority for men and women.15

Results fromTables 3 and 4can be used to illustrate the net impact of additional seniorityvia these two salary components. For example, consider a female faculty member with 5years of seniority. The first derivative of seniority with respect to COLA salary (based onthe coefficients reported inTable 3) indicates a seniority penalty of−$481.90.16 The resultsreported inTable 4 indicate that the increase in accumulated merit pay, associated withanother year of seniority for this faculty member, is $363.39.17 The net effect of anotheryear of seniority, based on these salary components, is−$118.51 (or,−$481.9 + $363.39).18

This example illustrates how cost-of-living adjustments that fail to keep pace with the

Page 12: Gender differences in faculty pay and faculty salary compression

176 K. Burke et al. / The Social Science Journal 42 (2005) 165–181

market rates contribute to a seniority penalty, even for faculty who receive a reward for theirproductivity.

The results from Model 3 with respect to rank and race are similar to those reported forModels 1 and 2. The coefficients for CHAIR indicate that faculty in these positions havemore accumulated merit pay, suggesting higher performance for these individuals. However,this claim cannot be supported at conventional levels of statistical significance. The resultswith respect to the entry-level CUPA salaries indicate that merit pay at U-PLAC is not influ-enced by changes in external market rates. The coefficients for TOP PERFORM do not differsignificantly (numerically or statistically) for female and male faculty, indicating an equal per-formance reward, regardless of gender. As mentioned above,Bellas et al. (2001)have arguedthat equal fixed-dollar merit allocations maintain absolute earnings differences. Our results areconsistent with this observation. The Chow and gender interaction tests do not indicate signif-icant differences between any of the female and male coefficients for Model 3. The similarityof female and male annual merit increases and the lack of a statistical difference in the slope ofthe merit-tenure profile suggests that gender equality in annual merit increases are associatedwith similar growth of accumulated merit pay.

Results of the decomposition analysis indicate that all (101.7%) of the gender gap in accu-mulated merit pay is explained by differences in faculty characteristics. The explained portionof a wage gap may be over 100% if the unexplained portion is negative. The lower regressionintercept for males contributes significantly to an unexplained portion that is less than zero.19

Consistent with the decomposition analysis of the other salary categories, rank and seniorityexplain a high percentage (103.2%) of the gender difference in merit.20 These results suggestthat, with an unbiased merit evaluation system, gender differences in accumulated merit willnarrow with greater equality in rank and years of service.

4. Conclusion and policy implications

Our findings underscore the importance of gender-neutral salary-setting practices in creatingsimilar salary growth rates for female and male faculty. In such an environment, equal access topromotion and retention appear to be the key to further reductions in relative pay differences.Furthermore, faculty salary growth rates are not always positive. Our results illustrate howannual cost-of-living increases that do not keep pace with market rates contribute to a senioritypenalty even on a campus where merit pay rises with seniority.

Finally, not all campuses share the salary-setting practices and other faculty characteristicspresent at U-PLAC. Hence, the source of gender inequities may vary from campus to campus.However, those institutions that share the recording distinction between merit and cost-of-living adjustments can use the techniques presented above to gain insight into the cause ofdisparity on a specific campus.

Notes

1. The amount of the gap that can be explained by differences in faculty attributes varies.For example,Barbezat (1991)attributes only 66% of the gap to differences in faculty

Page 13: Gender differences in faculty pay and faculty salary compression

K. Burke et al. / The Social Science Journal 42 (2005) 165–181 177

characteristics. SeeBellas et al. (2001)for a review of the factors contributing to genderdifferences in faculty pay.

2. Our data do not allow us to completely separate all of merit pay from COLA salary becauseone year’s merit is included in the next year’s base salary that will be adjusted by futurecost-of-living increases. The following example illustrates the extent to which we areable to isolate accumulated merit pay. Consider a faculty member who has completedtwo years of service at U-PLAC. At the end of each year this faculty member receiveda 3% cost-of-living adjustment, plus $500 in merit pay. Because the first year’s merit($500) becomes part of the base salary that is adjusted for the cost-of-living in the secondyear (3%× $500 = $15), the accumulated earnings effect of this faculty member’s meritis $1,015. Because of the manner in which merit increases are recorded at U-PLAC, weare unable to remove that amount of the merit increase that is affected by the cost-of-living adjustment. Consequently, recorded merit for this faculty member is $1,000, or$15 less than the total effect of previous merit awards. This $15 difference is included inCOLA salary. However, since cost-of-living adjustments at U-PLAC have historicallybeen quite low, ranging from 0 to 4% annually for the last 10 years, errors originatingfrom this source are expected to be low.

3. To illustrate the impact of fixed-dollar versus percentage increases on female–male earn-ings differences, consider a female who earns $30,000 and a male earning $40,000.The absolute wage gap is $10,000 ($40,000− $30,000) and the gap as a percent is 0.75($30,000/$40,000). If each receives a fixed-dollar increase of $800, absolute disparityremains $10,000 ($40,800− $30,800). However, the difference as a percent is now 0.755($30,800/$40,800). If each had received a 2% increase, the absolute difference would be$10,200 ($40,800− $30,600) with a constant percentage gap of 0.75 ($30,600/$40,800).

4. As is the case at many campuses, the negotiation of starting salaries is a two-step process.The chair negotiates this salary with the applicant and the administration. At U-PLACthere has historically been minimal power, within this negotiation framework, to increasesalaries above the amount originally budgeted for the position (negotiated increases rangeup to $1,500 above the original offer). In addition, cost-of-living adjustments at U-PLAChave ranged from 0 to 4% over the past ten years. Even in the extreme case of a newmale applicant bargaining for a starting salary that is $1,500 higher than a newly hiredfemale faculty member, a 4% cost of living increase would result in an annual increase,due to the starting salary increment, of $60.

5. These results are available from the authors upon request.6. We describe below the application ofOaxaca (1973)decomposition technique to the

model estimates. This technique requires that the male and female model specificationsbe identical. However, since there are two departments at U-PLAC that are either allmale or all female, we cannot meet this criteria if department dummy variables areused. We also estimated all of the salary equations with the percent female in each de-partment in place of CUPA SAL. The results of these estimates are similar to thosereported below. Results from this specification are available from the authors uponrequest.

7. The average merit award for the previous year is $292.48 (standard deviation of $409.15)for males and $331.25 for females (standard deviation of $431.44). The coefficients

Page 14: Gender differences in faculty pay and faculty salary compression

178 K. Burke et al. / The Social Science Journal 42 (2005) 165–181

for TOP PERFORM are $1,474 for men and $3,855 for women, based on Model 1.These results indicate that the earnings effect of TOP PERFORM is greater than theaverage merit increase these faculty received for their recent high productivity. Thisindicates that the measured effect of top performance is substantially larger than the awardgiven in a single year. This would be the case for top performers with a history of highproductivity.

8. This is the same technique used byBarbezat (1987)and Toutkoushian (1998). Oax-aca’s method “decomposes” the earnings gap into the portion of the gap that can beexplained by differences in female and male average characteristics and the portion thatis unexplained by these differences. The method can be expressed by the following:

. Wf = αf + βfSf (1)

. Wm = αm + βmSm (2)

where(1) and(2) are the estimates of female and male faculty salaries, respectively. Inthese equations, earnings (Wi ) depend only one variable,Si which is equal to years ofseniority at the institution. The intercepts and slope coefficients are represented by thetermsαi , βi . The difference in male and female salaries can be expressed as:

. Wm − Wf = (αm + βmSm) − (αf + βfSf ) (3)

or by,

. Wm − Wf = [(αm − αf ) + (βm − βf )Sf ] + βm(Sm − Sf ) (4)

Eq. (4) illustrates that the earnings gap can be explained by differences in the av-erage characteristics of men and women [βm(Sm −Sf )]. The unexplained portion[(αm − αf ) + (βm − βf )Sf ] is due to differences in regression intercepts and how anattribute is rewarded, or differences in slope coefficients. The unexplained portionis often attributed to discrimination which may have the effect of decreasing theearnings of women regardless of the level of seniority (lower intercepts), and ofnot rewarding female years of service as much as male seniority (different slopeterms).

In the Oaxaca method, the portion of the wage gap that can be attributed to discrimina-tion is based on the algebra of the estimated equations, not on the statistical significanceof the coefficients. Therefore, problems may arise with the interpretation of the unex-plained portion of the wage gap if the coefficients are not statistically significant fromzero for one group (say men), but are for another group (say women). This problem willalso extend to the interpretation of the explained portion of the wage gap if a coefficientfor men in not significantly different than zero. That is, the explained and unexplainedportions of the wage gap are difficult to calculate and to interpret if we think ofβm

andβf , from Eq.(4), as being equal to zero. In the results presented below, we observegender differences in statistical significance for coefficients for Ph.D., CUPA SAL andTOP PERFORM. However, from the perspective of the explained portion of the wagegaps examined here, these factors contribute very little to gender salary differences. Forexample, differences in Ph.D. attainment explain 0.0% of the gap. Furthermore, CUPA

Page 15: Gender differences in faculty pay and faculty salary compression

K. Burke et al. / The Social Science Journal 42 (2005) 165–181 179

SAL and TOP PERFROM explain from−0.057 to 0.38% of the gap (see data reportedin Tables 2–4where gender difference in significance levels are present). Therefore, in anumeric and in a statistically significance sense, these factors can be considered to playan insignificant role in gender earnings differences.

9. It is important to note that this unexplained portion may not accurately account fordiscrimination if relevant factors are omitted from the model. In so doing, an omittedvariable bias is introduced and the portion attributed to discrimination will be biased. Forexample, if women are more qualified with respect to the omitted factor(s), the extentof discrimination measured by the Oaxaca method will likely be underestimated. For afurther discussion of this problem seeGoldberger (1984).

10. Of the male sample, 6.8% are assistant professors (with a standard deviation of 0.25)while 26.8% of the female sample holds this rank (with an standard deviation of 0.448).The difference between these averages is significant at the .05 level.

11. The critical value for the Chow test for the estimate of total salary is less than 1. None ofthe coefficients from the gender interaction estimate approached statistical significance,based on two-tailed tests. The Chow and dummy interaction test for the other salarycategories yielded similar findings.

12. Oaxaca and Ransom (1999)find that the unexplained portion of a wage gap varies withthe choice of the reference category of dummy variables. However, the contributions ofsets of dummy variables to the explained portion of the wage gap is not dependent on thechoice of the reference category. Since we focus on the contribution of sets of dummyvariables, such as rank, the explained portion of the gap is not influenced by the selectionof reference categories.

13. Without measures of rank, the seniority variables pick up more of the explained difference.For example, in the model with controls for rank, the seniority variables explain 50.7%of the salary gap (the rank variables explain 41.8%). In the model without controls forrank, the seniority variables explain 87% of the difference in faculty salaries.

14. Moore et al. (1998)observe that the seniority penalty decreases in magnitude and instatistical significance as detailed measures of faculty performance are included in theestimate. With this approach, the observed disappearance of the seniority penalty may bedue to improved model specification, or due to the correlation between the productivitymeasures and seniority. Our test avoids this possible confusion by separating the earningscomponent that is linked to faculty productivity for a more direct test of the relationbetween seniority and accumulated merit pay.

15. If the distribution of merit pay on this campus is biased in favor of junior faculty, asopposed to a distribution based solely on performance, then the concave merit-seniorityprofile is consistent with Ransom’s monopsony hypothesis. Universities would offermore merit to retain more mobile junior faculty.

16. The first derivative of COLA salary with respect to years of seniority is:

.δCOLA salary

δseniority= −798.2 + 2(31.63) seniority

If seniority = 5 years, the slope, or incremental salary change due to another year ofservice is−$481.90.

Page 16: Gender differences in faculty pay and faculty salary compression

180 K. Burke et al. / The Social Science Journal 42 (2005) 165–181

17. The first derivative of accumulated merit with respect to years of seniority is:

.δmerit

δseniority= 468.49− 2(10.51) seniority

If seniority = 5 years, the slope, or incremental salary change due to another year ofservice is $363.39.

18. This example does not imply that faculty salaries actually decrease with seniority. Ratherthis exercise is based on the movements along the salary profiles as seniority increases.In this sense, the penalty associated with another year of seniority is a relative decreasecompared to less senior faculty. In addition, the net decrease of−118.51 due to thecombined merit and COLA salary effects compares to a decrease of−173.36, based onthe first derivative of total salary with respect to seniority for a female with 5 years ofservice. Thet-values for the linear seniority variables reported inTable 1suggest thatseniority is not significantly related to total salary. The implied flat total salary-seniorityprofile is likely due to the offsetting effects of merit and COLA salary effects.

19. The unexplained portion of the wage gap is indicated by the difference in the interceptsand by the differences in the slope coefficients. Using the notation from above, the portiondue to discrimination is: [(αm − αf ) + (βm − βf )Sf ]. The larger female intercept from themerit estimate (αF) would contribute to a negative unexplained portion.

20. The results of the Chow tests from all of the estimates suggest that the system of re-wards on this campus is fundamentally similar for female and male faculty with respectto any salary component. This result is not surprising given the uniform effect of thesalary-setting practices at U-PLAC. We have used the Oaxaca decomposition techniqueto determine the percent of the gender gap that can be explained by differences in averagefaculty characteristics and an unexplained portion that is due to differences in how thosecharacteristics are rewarded (indicated by numeric difference in regression coefficients).However, the Chow tests suggest that, in a statistical sense, the slope coefficients aresimilar for male and female faculty, implying that the unexplained portion of the wagegap is not statistically significant. Therefore, in a statistically significant sense, the dif-ferences in salaries on this campus are rooted in the significant differences in averagecharacteristics. Results reported inTable 1indicate that faculty on this campus differsignificantly with respect to years of seniority, the rank of assistant and full professor,the position as chair and race. Of these differences, our use of the Oaxaca method hasidentified rank and seniority as the key contributors of gender differences in salary.

References

Barbezat, D. (1987). Salary differences by sex in the academic labor market.Journal of Human Resources, 22,443–455.

Barbezat, D. (1989). Affirmative action in higher education: Have two decades altered salary differentials by sexand race?Research in Labor Economics, 10, 107–156.

Barbezat, D. (1991). Updating estimates for male-female salary differences in the academic labor market.EconomicLetters, 36, 191–195.

Page 17: Gender differences in faculty pay and faculty salary compression

K. Burke et al. / The Social Science Journal 42 (2005) 165–181 181

Becker, G. (1975).Human capital. New York: Columbia University Press.Bellas, Ritchey, M., Neal, P., & Parmer, P. (2001). Gender differences in the salaries and salary growth rates of

university faculty: An exploratory study.Sociological Perspectives, 44, 163–188.Bereman, N., & Lengnick-Hall, M. (1994). Pay compression at public universities: The business school experience.

Public Personnel Management, 23, 469–480.Brown, B., & Woodbury, S. (1998). Seniority, external labor market and faculty pay.The Quarterly Review of

Economics and Finance, 38, 772–780.Bok, D. (1993).The cost of talent. New York: The Free Press.Goldberger, A. (1984). Reverse regression and salary discrimination.Journal of Human Resources, 19, 293–318.Katz, D. (1978). Faculty salaries, promotions, and productivity at a large university.American Economic Review,

63, 469–477.Moore, W., Newman, R., & Turnbull, G. (1998). Do Academic salaries decline with seniority?Journal of Labor

Economics, 16, 352–366.Oaxaca, R. (1973). Male-female wage differentials in urban labor markets.International Economic Review, 14,

693–709.Oaxaca, R., & Ransom, M. (1999). Identification in detailed wage decompositions.The Review of Economics and

Statistics, 81, 154–157.Ransom, M. (1993). Seniority and monopsony in the academic labor market.American Economic Review, 83,

221–233.Reskin, B., Liddy, D., Haignere, L., & Frances, L. (1992, July–August). Salary-setting practices that unfairly

disadvantage women faculty.Academe, 32–35.Siegfried, J., & White, K. (1973). Financial rewards to research and teaching: A case study of academic economists.

American Economic Review, 63, 309–315.Toutkoushian, R. (1998). Sex matters less for younger faculty: Evidence of disaggregated pay disparities from 1988

to 1993 NCES surveys.Economics of Education Review, 17, 55–77.