The union effect on the earnings distribution in higher education in Ohio.

Benedict, Mary Ellen

Empirical work by economists has led to the general conclusion that unionization tends to compress the earnings distribution of workers across and within firms; this impact is often translated into an equalizing effect on earnings. However, from a theoretical perspective, the union effect on earnings inequality is ambiguous. When comparing firms in an industry, it may be that unions reduce earnings inequality by providing better salaries for workers previously on the low end of the distribution. On the other hand, unions may increase earnings inequality if only those workers who are in firms at the high end of the distribution are unionized. Another issue of ambiguity is how unions affect the salary inequality between worker groups. Within firms, unions often provide wage standardization rules that can reduce the inequality between worker groups, particularly by gender and race (Flaherty and Caniglia, 1992; Pfeffer and Ross, 1981).

This study analyzes the impact of unionization on the earnings distribution as it applies to schools of higher education, with an emphasis on gender differences. It investigates whether unions alter the earnings distribution in higher education, particularly between males and females as well as within each gender group. Given previous work on private-sector workers (Freeman, 1980, 1982; Hirsch, 1982; Flaherty and Caniglia, 1992), it would be expected that: (1) unionization has different effects on male and female earnings distributions; and (2) within the unionized schools, wage standardization reduces earnings disparity. Using micro-level data from seven Ohio public universities for the 1992-93 school year, this research analyzes the difference in faculty earnings distributions within each school and between union and nonunion institutions, and tests whether a union effect exists on the male-female earnings differential for faculty. The study uses standard methods of analysis: a variance of the natural logarithm of earnings comparison and multiple regression corrected for heteroskedasticity. It also has the advantage of using micro-level data to address the issues of earnings distribution, which can be obscured when using aggregate data.

An analysis of the union impact on the earnings distribution of these universities will provide new insight into gender differentials in academia. Most research on higher education salaries focuses on the union/nonunion salary differential (e.g., Freeman, 1978; Hu and Leslie, 1982; Barbezat, 1987; Kesselring, 1991) or gender differentials (e.g., Gray, 1990; Haberfeld and Shenhav, 1990) but little work has been done on the union effect on male/female faculty salary differentials. In addition, it would be interesting to see whether collective bargaining has the same effect in academic institutions as it does in the private sector where the net effect of unionization has been to decrease earnings disparity.

The following analysis indicates that the effect of unionization in higher education is indeed different than from that in the private sector - overall, unionization increases earnings dispersion among faculty, mainly due to the earnings disparity among unionized females. These results are described below.

The Relationship between Unionization and Earnings Inequality

As previously mentioned, a priori, it is not clear how unionization will affect earnings inequality across firms within an industry due to competing effects on the distribution of earnings. Because collective bargaining typically raises the wages and other benefits of workers, unionized firms tend to lower the quantity of labor demanded in their labor markets. This effect in turn forces workers unable to obtain a job in a unionized firm into lower-paying, nonunion firms. This outcome is usually called the "spillover" effect. At the same time, however, the "threat" effect takes place, and nonunion firms that wish to remain union-free may raise earnings in order to avoid unionization. Additionally, within firms, unions tend to have an equalizing effect - unionized firms tend to compress wages among worker subgroups due to the formalization of the wage structure and incentive structures for hourly and piece rate workers, and standardization of the promotion process. Ultimately, it is difficult to predict the net effect of unionization on earnings inequality overall: it depends on whether the spillover effect is greater than the other equalizing effects.

Empirical tests of the effect of unionization on the earnings distribution has been conducted mainly on private-sector wages. Freeman (1980, 1982) focused his analysis around wage standardization policies rather than the spillover and threat effects. Using establishment-level data obtained from the Bureau of Labor Statistics' Industry Wage Survey (1982), Freeman analyzed the variance of the natural log of earnings and used wage regressions. He found that unionism consistently reduced the earnings dispersion of workers within firms. He attributed the results to union wage policies, which tend to standardize wages within establishments and within industries.

In an earlier study, Freeman (1980) addressed whether the wage standardization practices of unions compress the dispersion of earnings of private-sector male workers. Using individual data from the Current Population Survey (CPS), Freeman found that unionization reduced wage inequality in two ways: (1) Workers covered by union contracts have much narrower wage dispersions than workers who are not associated with unions, which in turns lowers earnings inequality within highly unionized sectors; (2) The wage dispersion is much narrower between blue and white collar workers when comparing unionized and nonunionized firms. Later research has generally confirmed this result (Hirsch, 1982; Freeman and Medoff, 1984), even when investigating intra-race and inter-race differences (Pfeffer and Ross, 1981).

The inverse relationship between unionism and earnings inequality does not necessarily hold, however, when analyzing the wages of females. MacPherson and Stewart (1987) constructed a sample from the CPS in the same manner as did Freeman (1980) to analyze differences between male and female earnings dispersions in the private sector. They found that unionism reduces earnings inequality for males, but increases inequality for females. The authors suggested that this result may occur because nonunion females are more often in part-time jobs, while union females are not. Flaherty and Caniglia (1992) used aggregated 1980 Census data by state and found similar results.

Thus, while there is a leveling effect of unionism on the earnings of private-sector workers, it occurs predominately for males. This effect appears larger than any disequalizing effect on female earnings. Whether the same results occur with faculty earnings will be explored in the following sections.

Data on Seven Ohio Public Universities

The data used in this investigation consist of 4512 individual full-time faculty members from seven Ohio public universities for the 1992-93 academic year.(1) Of the seven schools, three were unionized by the fall of 1992, with one university voting for collective bargaining just prior to the year of analysis. Four universities were nonunion, although one of these had been in a collective bargaining election three years previously (and voted "no agent") and a second was in the early stages of an election campaign (and subsequently voted "no agent" in 1994).

The universities are similar in many respects. They are roughly comparable in the type of students who attend (a large percentage of students are regionally based) and in reputation, with an emphasis placed on teaching over research in each institution. Although individual schools can and do make many independent decisions and each has its own Board of Trustees, the Ohio public university system is ultimately governed by the Ohio Board of Regents and broadly-defined goals are mandated at the state level. For example, the university system recently underwent a review of current doctoral programs. The criteria used to evaluate individual programs and the final decision on which programs to terminate was made at the state level. In addition, six of the seven schools are classified as Category I (doctoral-degree granting) universities, as defined by the American Association of University Professors (AAUP). These six schools do not include Ohio State University, which is unique to all other public institutions in Ohio in terms of size and research reputation. The seventh school, which is unionized, is classified as Category II-A (comprehensive), as defined by AAUP.

Each observation included data on annual salary, age, gender, rank, years of service, tenure status, degree status, and department. Annual salary is defined as the annual 9-month salary for an individual and does not include fringe benefits.(2) Administrators and faculty from branch campuses were excluded. Appendix 1 contains the descriptive statistics for the final pool of faculty, and includes breakdowns by union status and by university. These estimates indicate that union and nonunion schools are very similar in several characteristics: Over 70% of the faculty in the seven universities are tenured and hold doctoral degrees, and the average faculty age for each school is about 47 years old. Average salaries also are very similar, and t-tests suggest no discernible differences between union and nonunion average annual salaries.(3)

There are also some slight differences among the schools. About 68% of the faculty have attained the rank of full or associate professor in both union and nonunionized universities, but there are slightly more full professors in the unionized schools (39% versus 35%). Further, the percent of female faculty in each school ranges from 21 to 29%; however, the average percent of female faculty is not very different when comparing union and nonunion status. Overall, the descriptive characteristics suggest that union and nonunion universities contain similar distributions of faculty.

The next section will explore the earnings distributions of union and nonunion schools more closely to examine whether differences exist.

The Earnings Distribution of Union and NonUnion Schools

The measurement used by Freeman (1980) - the standard deviation of the variance of log earnings - will be employed as a starting point of this investigation.4 Wage equations will then be used to discover whether systematic differences between union and nonunion schools may be associated with union status.

Table 1 presents the standard deviation of the natural log of annual salary, [Sigma], for all faculty, by union status and gender. The results indicate that unionized schools have a slightly larger variability in earnings than nonunion schools (.289 versus .279). The difference holds true when comparing unionized and nonunionized workers by gender. And, while all three differences in [Sigma] are statistically significant, the size of difference in dispersion between union and nonunion groups is small, particularly with males, where the difference in ln of earnings is only .004. Figure 1 further demonstrates that the salary distribution of union and nonunion faculty are very close, although there appears to be a slight trend for higher-paid unionized faculty to skew the distribution to the left. It appears then that although the difference in dispersion is small, that more highly-paid unionized faculty have a slight advantage in salaries than their nonunion counterparts.

As mentioned earlier, unionization is generally found to reduce earnings inequality between union [TABULAR DATA FOR TABLE 1 OMITTED] and nonunion sectors and within unionized settings. And, while unionized females earnings have been found to increase earnings inequality among working women across firms (MacPherson and Stewart, 1987; Flaherty and Caniglia, 1992), the effect is believed to be caused by unionized females who are situated at the high end of the earnings distribution. Nonunion females, who are more likely to be employed part-time, tend to be at the low end. In these data, both male and female faculty are on yearly contracts and generally work full-time. Therefore, these initial results are rather surprising because they suggest that unionization leads to slightly higher earnings inequality among all faculty in the sample, and women in particular.

We can consider several alternative explanations for the rise in earnings inequality with unionization. First, [Sigma] tends to weigh inequality more heavily at the lower end of the distribution than at the upper end when compared to other measures of income inequality, such as the Gini coefficient. Thus, the analysis places more emphasis on the differences in earnings between union and nonunion faculty who are on the lower end of the pay scales (typically non-tenure track employees and assistant professors) than on differences, for example, between tenured and nontenured professors. As a result, the measure used here tends to understate any equalization effect of a wage standardization policy that may result from unionization. Further, if women tend to be in the lower end of the distribution, as is the case here,(5) then the effect of women's earnings on faculty earnings inequality will be complicated by the above weighting scheme. A second reason may be related to the demographic characteristics of the faculty in the sample. Due to a large percentage of faculty who are tenured in unionized universities (around 70%), the majority of faculty may demand favorable wage policies for older, higher ranked members at the expense of untenured assistant professors and nontenure track faculty. If so, the earnings inequality within unionized schools will rise.(6) The comparison to nonunionized schools will be even greater if nonunion schools do not or cannot demand the same kind of preferences for senior faculty members.(7)

In order to investigate these alternative explanations, we must first delineate [Sigma] by union, gender and rank status to investigate earnings inequality within union and nonunion settings.

Differences in Earnings Dispersion

A. Within Union and Nonunion Settings by Gender. Table 2 presents [Sigma] by union and gender status and for each individual school to investigate whether differences in the earnings dispersion within unionized and nonunionized settings can be attributed to gender differences. In general, the addition of female salaries reduces the variance of the earnings distribution, except for one union school (School 2, the university classified as II-A, and where collective bargaining has existed for over twenty years) where female earnings tend to increase earnings dispersion. This general result may be largely attributed to women being heavily concentrated in certain occupations. MacPherson and Stewart (1987) found that women workers in the private sector were often aggregated into a few occupations, such as secretarial and clerical or nursing, and therefore the variance in female earnings tended to be smaller than that of males. In higher education, occupational categories would translate into academic disciplines, and it does appear that women in this sample are more highly concentrated: 25% of women faculty are in education and 28% are in humanities, while men are more evenly distributed across disciplines. Thus, it may be that the presence of female faculty lowers earnings dispersion within schools because they are crowded into disciplines; this market effect depresses the price of female labor and compresses the earnings distribution.

[TABULAR DATA FOR TABLE 2 OMITTED]

A second point that arises from a comparison of by school is that the largest variances of In earnings occur in the schools that were most recently involved in voting on faculty collective bargaining. Schools 3, 4, and 6 have the largest estimated earnings dispersion. Two of these schools recently voted on union representation, with one voting for unionization, and a third was in the process of a collective bargaining election during the 1992-93 school year. This result may be suggestive of an endogenous relationship between earnings disparity and unionization. In other words, it may be that faculty in a university with relatively large differences in earnings may consider unionization as a way to combat inequality. However, because earnings over time is needed to calculate a dynamic estimate of earnings dispersion, this study's result can only be suggestive of the endogeneity of earnings inequality and unionization.

Table 2 also suggests that a nonunionized setting leads to somewhat larger differences between the dispersion of salaries of males and females. The differences between [Sigma] for males and females are larger for nonunion universities, and although F-tests on union schools are not significant at the 5% level, all but one nonunion university demonstrates a significant difference in male/female salary dispersion.

To examine this result more closely, the variances of In earnings by union status have been further decomposed to see what share each factor has on the earnings dispersion. This decomposition can be written as:

[[Sigma].sup.2] = [Alpha]([[[Sigma].sup.2].sub.m]) + (1 - [Alpha]([[[Sigma].sup.2].sub.f]) +

[Alpha](1 - [Alpha]))[[W.sub.f] - [W.sub.m]).sup.2] (1)

where [Alpha] is the percent of males in a sector, [[[Sigma].sup.2].sub.j] is the variance of In earnings for each gender, and [([W.sub.f] - [W.sub.m]).sup.2] is the square of the difference in average In earnings of females and males. The first part of the equation estimates the male contribution to the variance in ln earnings, the second portion estimates the female contribution, and the final portion estimates the effect of male/female salary differences on earnings disparity. Table 3 presents these estimates for union and nonunion schools. As suspected, male salaries contribute largely to the dispersion in ln earnings for both sectors. However, within the union sector, female salaries make up about 22% of the inequality, while the gender differential contributes about 12% to the variance in ln earnings. Within nonunion schools, female [TABULAR DATA FOR TABLE 3 OMITTED] salaries contribute about 19% to the variance in ln earnings, and the salary differential contributes about 14% to overall disparity. Overall, the results indicate that the equalizing effect of female salaries is stronger in nonunion schools and may overcome the slightly disequalizing effect of the gender differential in these schools. This may be one reason why unionized universities exhibit slightly larger earnings disparity in comparison.

B. By Rank and Union Status. Table 4 presents an analysis of [Sigma] by rank and union status. Looking first across union and nonunion schools, we find that associate professors have a lower variance in ln earnings than all other ranks, regardless of union status. Further, the variance is larger for both assistant and associate unionized professors in comparison to their nonunion counterparts. However, full professors in nonunion schools have a larger variance than their unionized counterparts. F-tests that compare dispersion by rank and union status indicate that these differences are statistically significant, except for the instructor/lecturer group.

Within unionized universities, we find that the earnings of the higher ranked faculty decrease earnings inequality. Comparisons of the variance of In salaries of full and assistant professors, and tenured and nontenured professors, indicate no statistically significant differences between the higher and lower-ranked faculty in unionized schools, while there are significant differences among the nonunion faculty. In sum, the evidence suggests the salaries of the majority of faculty in unionized universities (the tenured, higher-ranked individuals) do not increase earnings disparity. Perhaps these senior workers use their collective status to make other type of gains. As suggested by other studies on higher education collective bargaining, senior faculty who tend to unionize show great concern for maintaining status and increasing the faculty role in governance (Bacharach, et. al, 1987). Alternatively, because the data lack information on other forms of compensation, it may be that senior faculty make gains through better pension and health insurance plans, factors that cannot be tested here.

Gender Salary Differentials by Union Status

The evidence above suggests that the equalizing effect of female salaries on the earnings distribution is not as strong in unionized schools as it is for nonunion schools. A regression analysis of the impact of collective bargaining on salary differentials between genders will shed further light on this result. It may be that unions help raise the earnings of higher-ranked female faculty in particular. If this is so, then the equalizing effect of female salaries within a unionized school will be dampened by the disequalizing effect of females on the higher end of the distribution getting larger wage gains when compared to the lower-ranked female faculty. Further, by controlling for individual effects on salaries, the effect of union status can be more clearly investigated.

Following the standard used in the labor economics literature, earnings are assumed to be distributed lognormally and are a function of the human capital and market explanatory variables. Separate equations are estimated for union and nonunion faculty to allow for possible differences in the effect of the explanatory variables, as well as differences in the variance in log earnings(8):

ln [W.sub.u] = [X.sub.u][[Beta].sub.u] + [[Epsilon].sub.u] (2)

ln [W.sub.n] = [X.sub.n][[Beta].sub.n] + [[Epsilon].sub.n] (3)

where the subscript indicates whether the observation is union or nonunion. In [W.sub.j] is a vector of annual salaries for individual faculty, [[Beta].sub.j] are partial coefficients to be estimated, and [[Epsilon].sub.j] are random error terms. The [X.sub.j] represent vectors of individual characteristics, and include: (1) measures of human capital investment (age, as a proxy for general experience, degree status, as a measure of educational investment, and years of service within an institution, as a measure of university-specific experience); (2) measures of individual productivity (rank and tenure); and, (3) a measure of labor market effects (broad discipline groupings, as indicated in Appendix 1). Because the earlier discussion indicated that occupational crowding of females into disciplines may affect the dispersion of earnings, the percentage of females in an individual's department is included to capture any occupational crowding effects on earnings. In addition, because one of the schools is not in the same AAUP category as the other six, a binary regressor is included to indicate if an individual works at the school classified as II-A (comprehensive). Finally, binary regressors for gender and union status are included in the specification.

Because the earlier analysis indicates that the variance of ln earnings varies by gender and union status, weighted-least squares (WLS) is employed to insure that the results include consistent standard errors for the estimated coeffcients. The estimated standard deviations of log earnings presented in Table 1 are used in the weighting scheme.(9)

Table 5 presents the estimates for Eqs. (2) and (3). A Chow test indicates that the differences between the union and nonunion coefficients are statistically significant.(10) Adjusted [R.sup.2] and F statistics suggest that a large degree of the variance in log earnings is explained by the equations. The first set of results in Column (1) indicates that the average annual salary of unionized faculty is not significantly different than the average salary of their nonunionized counterparts, ceteris paribus. Results from other studies are mixed concerning the union effect on salaries. While some positive salary effects of unionism was found empirically with 1970s data, more recent work has found little or no salary differential between union and nonunion faculty (Barbezat, 1989; Kesselring, 1991; Rees, 199(?)).

However, there are several important differences between union and nonunion schools. First, the returns to higher ranks (assistant, associate and full professor) are greater for unionized faculty. Unionized tenure-track faculty see anywhere from a .205 to a .567 increase in average ln earnings, which translates into as much as a 76% increase in average annual salary when compared to instructors and lecturers. In comparison, nonunionized faculty in tenure-track positions receive an increase from .194 to .504 when compared to that of non tenure-track faculty. This result is consistent with earlier empirical findings on the impact of unionization in higher education (Freeman, 1978; Barbezat, 1989) and could be due to the higher ranked faculty's negotiating better benefits for themselves. Second, it appears that the experience variables (age and years of service) have no relationship to earnings for unionized faculty, but statistically significant relationships exist within nonunion schools. The sign of the coefficients for years of service indicate that, ceteris paribus, the returns to years of service decrease at an increasing rate in nonunion schools, although the size of the effect is very small. This also is consistent with Barbezat (1989), who found a negative effect of seniority for nonunion faculty, but a positive effect of seniority (i.e., years of service) in unionized settings. These results are suggestive of a cohort effect. Because the model controls for rank, university-specific experience will likely reduce earnings as an individual stays with the school, particularly when comparing more recent hires to faculty employed in the same rank and university for many years.

TABLE 5.

WLS Regression Results
Dependent Variable = LN Salary

 All Schools Union Non-Union

Intercept 9.959(***)(a) 10.069(***) 9.899(***)
 (.055) (.088) (.071)

Asstprof 0.205(***) 0.205(***) 0.194(***)
 (.011) (.020) (.014)

Assocprof 0.324(***) 0.346(***) 0.290(***)
 (.014) (.024) (.018)

Full 0.542(***) 0.567(***) 0.504(***)
 (.014) (.025) (.018)

Age 0.010(***) 0.004 0.014(***)
 (.002) (.004) (.003)

Age(2) -0.0001(***) -0.00001 -0.0001(***)
 (.00002) (.00004) (.00003)

Gender -0.036(***) -0.027(***) -0.043(***)
 (.006) (.010) (.007)

Tenure 0.025(**) -0.004 0.054(***)
 (.010) (.015) (.013)

Doctorate 0.082(***) 0.072(***) 0.088(***)
 (.006) (.010) (.008)

Yrserv -.002 0.002 -0.005(***)
 (.001) (.002) (.002)

Yrserv(2) -0.0002(***) 0.0001 0.0002(***)
 (.00004) (.0001) (.00005)

Union 0.004
 (.005)

Business/Law(b) 0.208(***) 0.269(***) 0.173(***)
 (.009) (.015) (.012)

Social Science -0.012 0.005 -0.022(**)
 (.009) (.014) (.011)

Humanities -0.055(***) -0.018 -0.075(***)
 (.008) (.014) (.011)

Math/Comp Sci. 0.039(***) 0.050(***) 0.033**
 (.011) (.018) (.014)

Education 0.003 0.010 -0.0004
 (.010) (.016) (.013)

Engin./Tech 0.040(***) 0.058(***) 0.034(**)
 (.010) (.016) (.013)

Health/Hum Serv 0.109(***) 0.143(***) 0.075(***)
 (.015) (.022) (.020)

Percent Female -0.108(***) -0.111(***) -0.096(***)
 (.016) (.023) (.021)

Comprehensive -0.028(***) -0.031(***)
 (.008)

N 4512 1776 2736

Adj-[R.sup.2] .735 .748 .732

F 627.618 277.930 416.369

a Standard errors in parenthesis: *** = significant at [Alpha] = .01,
** = significant at [Alpha] = .05, * = significant at
[Alpha] = .10.

b omitted discipline (benchmark) is Natural Science.

Third, as with the experience variables, tenure attainment is unimportant in unionized settings, but increases average annual salary in nonunionized settings by 5.4%. The possession of a doctorate degree also has a higher return to those in nonunion schools. It appears that while rank has less of an impact on earnings in nonunion schools, tenure and degree status more than make up for the differences across union and nonunion status.

Fourth, the university status also has a statistically significant impact on annual ln salary. Those faculty working at the Category II-A (comprehensive) school are paid on average about 3% less than these at doctoral-degree granting institutions.

Differences in the gender coefficient also arise. While females earn less on average than similarly-situated males, those in unionized settings have slightly less gender disparity. Thus, although the earlier analysis suggests that the earnings distribution is not affected by female salaries in unionized settings, unionization tends to reduce the average differences in salaries between males and females. This finding supports earlier work by Smith (1992) who reviewed faculty average salary differences and found that although unionization does not eliminate salary differentials between genders, it does reduce the size of the differential.

A final set of differences involve the coefficients related to disciplines and the percentage of females in a department. The estimated coefficients indicate that unionized schools have statistically insignificant differences between traditionally female disciplines (Social Science, Humanities and Education) and the benchmark or omitted discipline (Natural Science). However, holding discipline constant, an increase in the percentage of females in a department has a larger negative effect on earnings for unionized faculty than for their nonunion counterparts (an 11.1% versus a 9.6% reduction in salary, for union and nonunion faculty, respectively). Thus, unionized schools generally appear to have more standardization of earnings across disciplines, but that effect is dampened when there are more females within a department or discipline.

These regression results support the evidence presented by the aggregated statistical analysis in the earlier section. It does not appear that the positive effect of unionism on earnings disparity is due to unions increasing average wage disparity between genders. In fact, collective bargaining appears to have a slight equalizing effect in this respect. Thus, it may be that the disparity arises within the unionized female subgroup, and consequently, a further examination of whether unionism increases wage disparity among female faculty is warranted. To investigate whether this supposition is correct, a WLS regression was constructed to include interaction terms between gender and rank. As shown in Table 6, these interaction terms are statistically significant.(11)

These results indicate that:

(1) When comparing women by union status, female faculty in unionized schools enjoy greater returns to rank, with assistant professors earning 25.1% more in earnings than the instructor/lecturer group, and associate and full professors earning 39.0% and 58.1% more, respectively. In nonunion schools, these returns are 16.2%, 27.6%, and 45.6% for assistant, associate and full professors, respectively (See Table 7). These differences are statistically significant at the 1% level as indicated by restricted F-test analysis.

TABLE 6.

WLS Regression Results
Dependent Variables = LN Salary

 Union Non-Union

Intercept 10.151(***)(a) 9.863(***)
 (.0901) (.072)

Asstprof 0.126(***) 0.238(***)
 (.043) (.020)

Assocprof 0.270(***) 0.321(***)
 (.035) (.022)

Fullprof 0.496(***) 0.543(***)
 (.036) (.023)

Age 0.0040 .014(***)
 (.004) (.003)

Age (2) -0.000005 -0.0001(***)
 (.00004) (.00003)

Gender -0.137(***) 0.017
 (.039) (.013)

Tenure -0.005 0.056(***)
 (.013) (.013)

Doctorate 0.072(***) 0.088(***)
 (.009) (.008)

Yrserv 0.002 -0.005(***)
 (.009) (.002)

Yrserv(2) 0.0001 0.0002(***)
 (.0001) (.00005)

Gender*Asstprof 0.125(***) -0.076(***)
 (.040) (.025)

Gender*Asscprof 0.120(***) -0.045(*)
 (.041) (.025)

Gender*Full 0.085(*) -0.087(***)
 (.042) (.027)

Business/Law 0.270(***) 0.174(***)
 (.015) (.012)

Social Science 0.004 -0.021(**)
 (.014) (.011)

Humanities -0.019 -0.073(***)
 (.014) (.011)

Math/Comp. Sci. 0.051(***) 0.033(**)
 (.018) (.014)

Education 0.008 0.002(**)
 (.016) (.016)

Engin./Tech 0.056(***) 0.035(**)
 (.016) (.013)

Health/Hum Serv 0.142(***) 0.074(***)
 (.022) (.020)

Percent Female -0.105(***) -0.098(***)
 (.023) (.021)

Comprehensive -0.032(***)
 (.008)

N 1776 2736

Adj-[R.sup.2] .749 .733

F 241.840 359.094

a Standard errors in parenthesis: *** = significant at
[Alpha] =.01, ** = significant at [Alpha] = .05, * = significant at
[Alpha] = .10.

b omitted discipline (benchmark) is Natural Science.

(2) While the female instructor/lecturer group earns about 14% less in earnings than similarly-situated males in unionized schools, the female and male non-tenure track faculty are virtually identical in average salary in nonunionized schools. However, nonunionized tenure track female professors earn on average lower returns to rank when compared to their male counterparts, and these differences are larger than that in unionized schools. (See Table 7). The unionized assistant and associate professor ranks exhibit the smallest differences in male and female average salaries (1.2% and 1.7%, respectively), while the full professor rank has the largest differences for all tenure-track ranks (5.2% and 7.0% for union and nonunion schools, respectively).

These regression results increase our understanding of the union effect on faculty earnings dispersion. It appears as though the larger disparity at union schools is in part due to larger returns made by senior female faculty, particularly full professors. This union effect tends to reduce the male/female wage disparity at higher ranks at the expense of non-tenure track females. It may be that the union promotes wage standardization policies for a subgroup of workers, in this case, tenure-track professors.

Another explanation may be that the schools most likely associated with collective bargaining in higher education promote gender equality. The three major unions in academia are the American Association of University Professors (AAUP), the American Federation of Teachers (AFT) and the National Education Association (NEA) (Bacharach, er. al, 1987). AAUP publicly supports pay equity issues, and the last two have extensive experience in organizing public schools where female workers predominate. And, collective bargaining contracts in higher educated have addressed pay equity across disciplines and comparable worth programs (Bacharach, 1987).

[TABULAR DATA FOR TABLE 7 OMITTED]

Conclusion

Using data on seven Ohio public universities, this study investigates whether collective bargaining decreases the earnings distribution of faculty. Research in the private sector indicates an inverse relationship between earnings disparity and unionism, but faculty tend to be in a unique occupation that may not lead to similar effects. Indeed, this study suggests that unionism increases overall salary dispersion slightly. This increase appears to occur in part due to the gains senior female faculty make in unionized settings when compared to non tenure-track females and not increases in male/female salary disparity. In fact, in this sample, [TABULAR DATA FOR APPENDIX 1 OMITTED] unions decrease male/female wage disparity at higher ranks, but at the instructor/lecturer level, the disparity between males and females is greater than in nonunionized schools.

These results should be viewed with some caution given the nature of the sample. The advantage of these data is that the observations are by individuals and therefore avoid the problems of aggregation bias seen in studies using establishment data. However, using a nonrandom sample of seven schools limits the ability to generalize from the study to other universities where collective bargaining in higher education may be governed by different legal structures than what occurs in Ohio or where school quality may vary.

Department of Economics, Bowling Green State University, Bowling Green, OH 43403. E-Mail: mbenedi@opie.bgsu.edu. Many thanks to my colleagues attending the session at the American Economic Association in 1995 and the BGSU Economics Colloquium Series in 1994. Thanks also to an anonymous referee, Professors Linda Babcock, John Hoag, Neil Browne and Lisa Wilder for their comments, to Professor Michael Magura of the University of Toledo for sharing the data on universities with me, and to Kelly Shaffer for her research and word processing assistance.

NOTES

1. As per my agreement with my data contacts, I am unable to list the names of the universities used in this analysis. However, I provide a general description of the group of schools later on in the text of the paper.

2. When possible, data were converted into 9-month salaries. For universities that did not provide such information, it is assumed that the data were for 9-month contracts. Given the descriptive statistics of each university, this appears to be a reasonable assumption in that averages and ranges are quite similar across the seven schools. Further, fringe benefits are not added to the analysis. An examination of fringe benefits by union status would likely add insight into this investigation because unions typically create more opportunities for such benefits and therefore would change the distribution of total compensation. Unfortunately, fringe benefit data were not available for individual faculty and therefore the focus of this analysis is on the dispersion of salaries.

3. A t-test on the differences between average union and nonunion salaries produced a t-statistic of .345, which does not exceed conventional levels of significance.

4. Freeman (1980) and MacPherson and Stewart (1987) focus on the analysis of the variance of In earnings and wage equations to investigate the union effect on the earnings distribution. Other researchers, such as Flaherty and Caniglia (1992), use Gini coefficients and quintile shares as part of their analysis.

5. Average salaries by gender, rank, and school, indicate that women faculty in this sample typically get paid less than men. Details are available upon request.

6. Merit increases will increase disparity, too. However, it is not likely that unionized universities use a merit system that would yield larger income dispersion than their nonunionized counterparts.

7. A third explanation may be that the spillover effect dominates all other equalizing effects of unionization. However, the data here do not indicate that spillover effect exists. Salaries are comparable across schools and faculty are distributed similarly across union and nonunion settings.

8. Empirical research on union-nonunion earnings differentials has suggested that the correct method for estimating the differential involves a switching regression model that controls for union endogenity. However. in his comprehensive review of the union effect on labor markets, Lewis (1986) suggests that this model, while theoretically appealing, has not been supported by empirical evidence. In the case of faculty salaries, Barbezat (1989) and Kesselring (1991) include such a model in their analyses and do not find empirical support for it. Therefore, I use separate ordinary least squares regression models for union and nonunion faculty salary estimation.

9. The model assumes a homoskedastic error term among disciplines. To the degree that this is not the case, the weighting scheme may be more complex then what is used in the analysis.

10. The Chow test yielded an F-statistic of 5.166, which is significant at the 1% level of significance.

11. The interaction terms were jointly significant as well. A restricted F-test yielded a statistic of 4.86 for the nonunion sector and 4.09 for the union sector, both of which are at significant at the 1% level of significance.

REFERENCES

Bacharach, Samuel B., Timothy P. Schmidle, and Scott C. Bauer (1987). "Higher education," in Collective Bargaining in American Industry: Contemporary Perspectives and Future Decisions (D. Lipskey and C. Donn, eds.), 223-264, Lexington Books, Massachusetts.

Barbezat, Debra (1989). "The effect of collective bargaining on salaries in higher education," Industrial and Labor Relations Review 42(3), 443-455.

Flaherty, Sean and Alan Caniglia (1992). "The relative effects of unionism on the earnings distributions of women and men," Industrial Relations 31(2), 382-393.

Freeman, Richard B. (1978). "Should we organize? The effects of faculty unionism on academic compensation," Working Paper no. 301, National Bureau of Economic Research.

-----, (1980). "Unionism and the dispersion of wages," Industrial and Labor Relations Review 34(1), 3-23.

-----, (1982). "Union wage practices and wage dispersion within establishments," Industrial and Labor Relations Review 36(1), 3-21.

Freeman, Richard B. and James L. Medoff(1984). What Do Unions Do?, Basic Books, New York.

Gray, Mary W. (1990). "Achieving pay equity on campus," in AAUP Higher Education Salary Evaluation Kit (E. Scott, ed.), American Association of University Professors.

Haberfeld, Yitchak and Yehouda Shenhav (1990). "Are women and blacks closing the gap? Salary discrimination in American science during the 1970s and 1980s," Industrial and Labor Relations Review 44(1), 68-82.

Hemmasi, Masoud and Lee A. Graf (1993). "Determinants of faculty voting behaviour in union representation elections: a multivariate model," Journal of Management 19(1), 13-32.

Hirsch, Barry (1982). "The Interindustry structure of unionism, earnings, and earnings dispersion," Industrial and Labor Relations Review 36(1), 22-39.

Hu, T-W and L. L. Leslie (1982). "The effects of collective bargaining on faculty salaries and compensation," Applied Economics 14, 269-277.

Kesselring, R. G. (1991). "The economic effects of faculty unions," Journal of Labor Research 12(1), 61-72.

Lewis, Gregg (1986). Union Relative Wage Effects. The University of Chicago Press, Chicago.

MacPherson, David A. and James B. Stewart (1987). "Unionism and the dispersion of wages among blue-collar women," Journal of Labor Research 8(4), 395-405.

Pfeffer, Jeffrey and Jerry Ross (1981). "Unionization and income inequality," Industrial Relations 20(3), 271-285.

Rees, Daniel. 1993. "The Effect of Unionization on Faculty Salaries and Compensation: Estimates from the 1980's." Journal of Labor Research 14(4): 399-422.

Smith, Teresa (1992). "The impact of university faculty unionization on the male-female wage differential," Journal of Collective Negotiations 21(2), 101-110.