Does raiding explain the negative returns to faculty seniority?

Bratsberg, Bernt ; Ragan, James F., Jr. ; Warren, John T. 等

I. INTRODUCTION

Economists have studied various aspects of faculty mobility. Research shows that turnover is greater among nonunion faculty (Rees 1994); female faculty average slightly more employers than male faculty (Barbezat and Hughes 2001; Ginther and Kahn 2004) (1); the responsiveness of turnover to relative pay diminishes with rank (Ehrenberg, Kasper, and Rees 1991); and research productivity enhances prospects for upward mobility (Ault, Rutman, and Stevenson 1979, 1982; Coupe, Smeets, and Warzynski 2005). Still, questions remain unanswered: Where do faculty go when they leave a university? How does faculty turnover vary with time at a university? What percentage of faculty move upward and what percentage move downward because of tenure denials or the prospect of being turned down for tenure? The present article contributes to our understanding of these issues, but its primary objective was to use data on faculty mobility to provide insights on another issue: returns to academic seniority.

The negative returns to seniority that many studies find are commonly attributed to one of two factors. (2) The first view is that negative returns reflect monopsony power on the part of universities. Faculty with high mobility costs acquire high seniority and, because they are unlikely to move, they are paid less than equally productive faculty that employers view as more likely to leave. A competing explanation is that highly productive faculty are "raided." According to this story, returns to seniority are biased negatively because faculty with high seniority tend to have low unmeasured productivity. The best faculty are bid away, moving on to better jobs. Those who remain are the faculty without outside offers, and the negative returns estimated for senior faculty actually reflect a penalty for low unmeasured productivity rather than a true penalty for seniority.

The raiding story has several implications. First, estimated returns to seniority should become less negative the more fully the productivity is measured. Second, estimated returns should be less negative once we control for faculty who are bid away. Finally, and of particular importance, estimated returns to seniority should not be negative for faculty who are bid away, the faculty for whom unmeasured productivity is presumably high. With the right data set, these hypotheses can be tested. Rejection of the raiding hypothesis would provide tacit evidence of monoposony in academic markets. (3)

This article contributes to the literature in several ways. It is the first research to directly test the raiding hypothesis. It provides a detailed look at faculty mobility in one segment of the academic labor market. It extends and improves Bratsberg, Ragan, and Warren (2003) in the following ways: it expands the sample to cover three decades of faculty pay records; it includes additional controls for productivity: number and types of books, recent and prospective articles, and so forth, resulting in the most complete specification of research productivity to date; and it allows for heterogeneity in the seniority penalty via PhD pedigree and direction of move. Finally, it tests for the possibility that the academic job market does a better job of matching faculty and universities in recent years, as might be expected because of the information revolution.

The sample consists of five public, PhD-granting universities in the Midwest. We track all tenure-track economics faculty employed in these universities over a 30-yr period. We measure their productivity, their pay, and their mobility. As such, the study provides a rich and detailed investigation of one subset of economics programs. It is possible that mobility may be greater in urban areas, where faculty can more readily obtain alternative employment without relocating; and prior research indicates that pay structures depend on the university's stature and its relative emphasis on research and teaching. (4) Therefore, the results of this or any other study of specific universities cannot be generalized across institutions. Even so, detailed evidence on one segment of the academic labor market can provide new and important insights on faculty mobility, pay structures, returns to seniority, and, of particular interest, whether the data support the raiding story.

II. ACADEMIC PAY

For personal and scholarly reasons, economists are interested in the process through which pay of economics faculty is determined. A long literature has evolved, but the primary focus of recent research has been the relationship between seniority (years at the current academic institution) and pay. Many studies have found that, among nonunion faculty at research universities, pay is inversely related to seniority (Barbezat 1989; Brown and Woodbury 1998; Gordon, Morton, and Braden 1974; Ransom 1993). (5)

Moore, Newman, and Turnbull (1998) point out that most studies provide few if any measures of research output and, prior to their work, controlling for quality of research was rare. (6) Picking up on the raiding theme, they find that, for their sample of 142 associate and full professors in academic year 1993/1994, the coefficients of seniority and experience become insignificant at the 5% level once controls are added for quantity and quality of research (though the coefficient of seniority remains negative and, in one of two specifications, is significant at the 10% level). But in later research that adds assistant professors to the original data set (Moore, Newman, and Terrell 2007), the coefficient of seniority again emerges as negative and significant. (7)

Bratsberg, Ragan, and Warren (2003) study 176 faculty over a 21-yr period and find strong evidence of negative returns to seniority regardless of specification. Arguing that match quality imparts a positive bias to estimated returns to seniority, they then reestimate their model using the two-step estimator of Topel (1991). As predicted, the estimated seniority penalty increases once match quality is accounted for. At 15 yr of seniority, the estimated penalty is 14.6% of salary.

Why would returns to seniority be negative? Why would a newly hired faculty member earn more than someone with the same experience and a comparable research record? One argument is that, given difficulty in measuring faculty productivity, the two faculty may not really be comparable. Perhaps the faculty member with higher seniority has lower unmeasured productivity. This interpretation is consistent with the models of Harris and Holmstrom (1982) and Lazear (1986). The thesis is that faculty potential is difficult to determine at the time of the hire but is revealed as time passes. Faculty who turn out to be highly productive are raided by other institutions, leaving behind faculty with low unmeasured productivity.

According to the raiding story, seniority is not actually penalized. Instead, the negative returns that are estimated for seniority capture the inverse correlation between seniority and unmeasured productivity. As better measures of productivity are added, the estimated coefficient of seniority should become less negative. The estimated seniority penalty should shrink further when we control for faculty who are bid away. Finally, there is no reason that faculty who are raided should be penalized for their time with the university. If faculty who move to better or comparable jobs are subject to the same penalty as faculty who remain with the university, the reason for the negative seniority profile is not lower unmeasured productivity of those who remain behind. The answer must lie elsewhere.

A competing explanation is that the penalty to seniority is real and reflects monopsony power on the part of nonunion universities. Ransom (1993) and Boal and Ransom (1997) present formal models of monopsony that are based on differential mobility costs of faculty. Faculty who are reluctant to move, whether because of family or other considerations, fall behind otherwise similar faculty who are less averse to moving. Faculty with high mobility costs tend to achieve higher seniority and, because they are unlikely to leave, employers take advantage of the situation when setting pay.

This view is consistent with the experience of Ronald Ehrenberg who is not only a respected labor economist but also a former key administrator at Cornell University. Ehrenberg (1999, 107) writes that "our deans who allocate salary funds are ... exploiting the monopsony power that they have over the least mobile faculty." In contrast, other researchers (e.g., Monks and Robinson 2001; Moore, Newman, and Turnbull 1998) favor the raiding explanation. Because there has not yet been any formal test of raiding, alternative interpretations are possible. The data of the present study make it possible to test for raiding in the universities being studied.

III. FACULTY MOBILITY IN FIVE UNIVERSITIES

The data underlying this study come from the economics departments of five Midwestern universities. The schools are Iowa State University, Kansas State University, University of Kansas, University of Missouri, and University of Nebraska. Based on research output, the universities frequently are ranked between 50 and 100 or thereabouts. (8) All five universities offer doctoral degrees in economics and are located in communities whose populations range from 50,000 to 230,000. As such, faculty who leave these universities for alternative employment within the economics profession generally must relocate.

For each university, we collected annual pay records of all tenure-track faculty with at least a 0.5 appointment in the department of economics. Over the 30 years studied, 1975/1976 through 2004/2005, we collected data for 238 faculty. (9) Given the panel nature of the data, our sample has 2,698 observations of faculty salary. For each faculty member, we also gathered detailed information on administrative appointments, annual research productivity, date of doctoral degree, institution conferring the degree, date of initial employment in the department, and so forth.

Forty-five percent of the faculty worked elsewhere before joining our sample. Their postdoctoral experience at the time of hire ranged from 1 to 30 years, with a mean of 5.93 years. Whenever faculty left one of the universities for another job, we determined the subsequent employer. (10) Unfortunately, we do not know salaries except while the person is in our sample (though we do know the research output of faculty both before and after they are in our sample).

For faculty moving out of the sample, Table 1 shows the reason for leaving, and for those who continued to work, the type of subsequent employment. For faculty taking positions elsewhere, Figure 1 indicates whether the move was upward, downward, or horizontal. About two-thirds of the movers took another academic position. To determine the direction of academic moves, research rankings of the new and former departments were compared. Because rankings change over time, we used rankings at the time of the move. (11)

When the new university was ranked at least 20 positions higher than the university from which the faculty member moved, the move was classified as upward. When the new university was ranked at least 20 positions lower, the move was classified as downward. When the two universities were within 20 positions of one another, the move was considered lateral. For moves to another unit in the same university (another department, a research institute, or administration), the move was classified as lateral unless we had information that the move was involuntary. All moves to business and local government were labeled downward. For jobs with the federal government or a nonacademic foreign institution, moves were considered downward or lateral based on the prestige of the new position. For example, a move to the board of governors was considered lateral, and a move to a minor research institute downward. Whenever a university in our sample was ranked in the top 70, moves to government or nonacademic jobs were classified as downward. (12)

Based on these classification rules, 23% of the moves were upward, 11% were lateral, and 66% were downward. (13) Assuming that faculty who are raided because of their high productivity move upward or laterally, no more than one-third of leavers appear to have been "bid away." For the 100 faculty who moved (42% of the sample), Figure 2 shows the length of time the faculty stayed at the university. The most common durations were 4, 5, and 6 years. For faculty who left after 4-7 years, we looked at direction of move. if the move for these faculty was downward, the move was labeled a "tenure denial," (14) though some readers may prefer milder terminology. Surely, some of the moves were in anticipation of being denied tenure rather than formal termination, and other motivations are possible. But given the timing and direction of the moves, our measure can be viewed as a proxy for negative outcomes of the tenure process. Upward moves occurred anywhere between 2 and 23 years but were most frequent in the 5- to 7-years period, indicating that labor market activity is most prevalent in the period immediately preceding the tenure decision.

IV. EMPIRICAL SPECIFICATION OF THE WAGE EQUATION

Empirical models of academic pay are based on a human capital specification in which the log of a faculty member's pay depends on seniority, experience (age when experience is unavailable), gender, and other control variables. This is the approach we adopt. In particular, we estimate the following equation:

(1) ln [W.sub.ijt] = [[alpha].sub.1] [Sen.sub.it] + [[alpha].sub.2] [Sen.sup.2.sub.it] + [[alpha].sub.3] [Exp.sub.it] + [[alpha].sub.4] [Exp.sup.2.sub.it] + [beta][X.sub.it] + [[mu].sub.i] + [[epsilon].sub.ijt],

where [W.sub.ijt] is the 9-mo salary of faculty member i at university j in year t, Sen and Exp are years of seniority and experience, X is the vector of control variables, [[mu].sub.i] is an individual-specific term that captures unmeasured productivity (estimated as a random effect), (15) and [[epsilon].sub.ijt], represents white noise. (16) The control variables include gender (Female), current service as department chair (Chair), number of years the person has served as chair (Years Chair), past service as chair (Past Chair), past service as dean (Past Dean), number of completed dissertations for which the faculty member served as major professor (Advisees), academic year, and institution.

[FIGURE 2 OMITTED]

Also included is a rich array of variables designed to capture research productivity. First is Articles, the number of articles in journals listed in EconLit. (17) We control for quality of articles in several ways. First, we divide articles into four tiers. The first tier consists of articles published in top 10 journals, as determined by Moore, Newman, and Turnbull (1998). (18) The second tier comprised the 26 journals ranked by Scott and Mitias (1996) as being in the top 36 that are not already included in the top tier. Tier 3 is made up of journals that were rated by department chairs as falling in the top 50 (Enomoto and Ghosh 1993) and that are not listed in tiers 1 or 2 plus 3 other journals that often are ranked highly: Journal of Economic Perspectives, European Economic Review, and Economic History Review. (19) Tier 4, the reference category, consists of articles in all remaining journals listed in EconLit.

In one formulation, articles in the top three tiers are converted to American Economic Review-equivalent pages and deflated by the number of authors. These variables are labeled Pages Cat1, Pages Cat2, and Pages Cat3. In another version of the model, the variables are adjusted for neither page count nor number of authors. Instead, we simply count number of articles in each tier (Articles Cat1, ..., Articles Cat3).

Quality is also proxied by a quadratic specification of the number of times a faculty member's research is cited in the Social Science Citation Index, which is a component of the Web of Science. The count is number of publications to cite the faculty member, meaning that the faculty member gets one hit whether one or more articles appear in the bibliography.

In addition, we tabulate the number and type of books a faculty member has authored. We have separate tallies for book chapters (Book Chap), research books (Book Res), edited books (Book Ed), and textbooks (Book Text). (20) To allow for differential returns to subsequent editions of textbooks, we also count the number of revisions of the original textbook (Text Revis). Research books and book chapters are more closely related to scholarship than textbooks, but widely read textbooks may enhance the visibility of the department to a greater extent than research books, making it difficult to predict relative returns to the different categories of books. Because most research on pay has ignored books entirely, and those studies that do include books do not differentiate among the various categories of books, prior research does not offer guidance on the size of the relative parameters. (21)

Prior research on faculty pay has focused exclusively on past output, but it is possible that department chairs and evaluation committees consider prospective research flows. Indeed, the faculty member's evaluation portfolio (containing working papers, letters of acceptances, letters inviting revisions and resubmissions, and the like) provides a good basis for predicting publications in the near future. Therefore, in one version of the model, we include not just articles published through the end of the year in which a faculty member is evaluated but also publications in the subsequent 3 years. The variables capturing future publications are Prospective Articles, Prospective Pages Cat1, Prospective Pages Cat2, and Prospective Pages Cat3. The relative returns to prospective publications may be greater than or less than that of past research, depending on a department's faculty evaluation procedures.

In this more fully specified model, we also allow for differential (added) returns to articles published in the past 5 years: Recent Articles, Recent Pages Cat1, Recent Pages Cat2, and Recent Pages Cat3. If senior faculty have published little in recent years and if older publications are weighted less than recent publications, it is possible that the seniority penalty picked up in previous research actually captures a penalty to declining output flows of senior faculty. (22)

Some versions of the model also control for pedigree of the institution from which the faculty member graduated. Based on the same departmental rankings used to indicate whether a move was upward, downward, or horizontal, doctoral-granting institutions were classified as Top 10 if the university was rated as having one of the 10 best doctoral programs in economics at the time the degree was conferred. Siegfried and Stock (1999, 2004) find that graduates of top-tier universities earn more than other graduates, but presumably, the primary reason for the premium is the greater ability of graduates from elite programs to find employment in higher rated and higher paying departments. What we want to determine is whether, for a given academic institution, graduates from top programs come in at a higher salary or experience faster growth in compensation (a lower penalty to seniority).

Faculty from the top programs may be more marketable than other faculty. Their degree may be more likely to generate subsequent job interviews, and their network may also give them an edge in their job search. Deans may be willing to pay more to keep someone from a top program (to keep a Harvard or Chicago affiliation on the department's Web page). If faculty from top programs are perceived as more marketable, employers will be able to exert less monopsony power against such faculty. Faculty from top programs may also experience faster wage growth if, despite the large set of control variables, faculty from top programs have higher unmeasured productivity than faculty from lower rated institutions.

Finally, and of special interest, some specifications of the model allow for differential effects for three classifications of faculty: faculty who may have been raided--those who move upward or laterally (Bid Away), faculty who apparently were denied tenure (Denial), and other faculty who move (Other Move). Although our focus is on the first category of movers, we predict that returns to seniority will be most negative for faculty who are denied tenure, a proposition not previously tested. A university is not likely to worry about losing faculty whose productivity falls short of tenure standards, so why put much of a department's salary-increment pool on faculty who are on their way out?

Table 2 summarizes values of the key variables used in the analysis, and Table 3 provides sample means separately for nonmovers, faculty who are bid away, tenure denials, and other movers. The latter table shows that the faculty who are bid away tend to be highly productive (consistent with the findings of Ault, Rutman, and Stevenson 1979, 1982; Coupe, Smeets, and Warzynski 2005). They have more articles, prospective articles, and citations than any other group. Although nonmovers have more textbooks and pages in ranked journals, the average experience is twice as high for non-movers as for faculty who are bid away. At the other end of the spectrum, tenure denials have fewer recent articles and prospective articles and fewer recent pages in ranked journals than any other group.

Finally, females are disproportionately concentrated in the bid away and denial categories. Although we are reluctant to make too much of this point, given the small number of females in the sample (30 individuals, 193 observations), females are more likely to move out of our sample (both upward and downward) than males. Higher mobility of female faculty has been documented previously (e.g., Barbezat and Hughes 2001; Ginther and Kahn 2004). At least for this type of Institution, is that females are more likely to be bid away (especially in recent years) and also more likely to be denied tenure. (23)

V. EMPIRICAL ESTIMATES

A. Returns to Seniority and Productivity

In the stripped-down version of the model (Table 4, column 1), we find negative returns to seniority, positive but diminishing returns to experience, and no evidence of pay differences by gender. When controls are added for administrative experience and graduate advising (column 2), we see that these activities appear to be rewarded highly, but the estimated penalty for seniority is virtually unchanged--about 12.7% at 10 years of seniority. (24) Next, we include research productivity, finding that faculty are compensated for both quantity and quality of research (column 3). An unranked article is estimated to increase pay by 0.9%. The added returns to a 10-page, sole-authored article are 3.0% if published in the American Economic Review (or other tier 1 journal), 1.0% if in a tier 2 journal, and 0.6% if in a tier 3 journal. Faculty are compensated further if the article is cited.

Faculty are also rewarded for writing books and book chapters. Estimated returns are greatest for textbooks--4.3% for a first edition and 2.0% for each subsequent edition. Research books increase pay by an estimated 2.0%, and the payoffs for editing a book or writing a book chapter are 1.3% and 1.5%, respectively. The higher rewards for textbooks may be because of their greater visibility. (25) Alternatively, if faculty who write textbooks tend to be better teachers or spend more time with students, part of the textbook premium may reflect compensation for these other activities.

Interestingly, the coefficient of Advisees, which indicated a reward of 1.0% per advisee in column 2, becomes zero and insignificant once we control for research productivity. The implication is that graduate advising is not rewarded directly. Students are drawn to major professors with strong research records, and when research productivity is not accounted for in the regression, number of doctoral dissertations supervised serves as a proxy for research productivity. But when both sets of variables are included, we see that it is actually the higher research output of advisors that explains their higher pay.

The addition of research variables (column 3) reduces the penalty to seniority by approximately 20%, which is consistent with the raiding story. Therefore, there is evidence of a negative correlation between productivity and seniority. As Moore, Newman, and Turnbull argue, failure to account for research productivity does negatively bias estimated returns to seniority. But of greater relevance, the estimated penalty to seniority remains sizable and highly significant even when we control richly for quantity and quality of research. Our interpretation of these results is that, at least for our sample, the penalty for seniority is real.

We also estimated faculty pay based on an alternative specification of the research variables. Rather than adjusting for number of authors and pages, we simply count the number of articles that a faculty member has in each of the four journal tiers. We estimate that the return for publishing in a tier 1 journal is 2.7% and that returns are 1.9, 1.4, and 0.7%, respectively, for journals classified as tiers 2, 3, and 4. These results demonstrate that faculty pay depends more strongly on quality than quantity of publications and further suggest that there is a recognizable hierarchy of journals in the economics profession.

Column 4 provides estimates for the specification that includes impending publications and allows for differential returns to recent articles (a positive coefficient for a "recent" publication designates the added premium for articles published within the past 5 years). Returns to articles in ranked journals are modestly higher if published recently. In particular, for a tier 1 journal, the reward to a sole-authored, 10-page article is 0.6 percentage point higher if published in the past 5 years. (26) For tiers 2 and 3 journals, respectively, the added premiums are 0.5 and 0.1 percentage points. In contrast, for unranked journals, returns appear to be 0.3 percentage points higher for articles published more than 5 years ago. Publications in unranked journals are valued, but apparently, a portion of the reward is received with considerable delay.

The coefficients of all four variables capturing prospective articles (those to be published in the next 3 years) are positive, and two of the coefficients are statistically significant. But in all cases, returns are lower for prospective articles than for published articles and that is true for articles published more than 5 years ago as well as for recent articles. Although faculty can expect to be rewarded for future articles, those rewards will come primarily after the articles are in print; pay depends more strongly on past publications than on impending publications.

What really matters, however, is that including variables to capture prospective publications and allowing for lags does not appreciably affect estimated returns to seniority. The lower pay of senior faculty is not the consequence of differences in impending research flows or of a tendency to have dated publications whose value has depreciated over time. Based on the specification of column 3, the estimated penalty for 10 years of seniority is 9.8% of pay. When we add prospective articles and allow for differential returns to articles published in the past 5 years and those published earlier (column 4), the estimated penalty is 9.3%. Finally, when we try a specification that allows for differential returns to articles published in the past 5 years, the previous 5 years, and more than 10 yr ago and allows for differences in returns to both number of articles published by quality level and pages published by quality level, the estimated penalty to 10 years of tenure is 9.2%. As these numbers indicate, altering the specification of the productivity variables has only a trivial effect on the estimated penalty to seniority. (27) The lower pay of senior faculty cannot be explained in terms of different patterns of publishing.

B. Returns to Seniority by Faculty Mobility

Next, we allow for differential returns to seniority of faculty based on whether they move or stay and, if they leave the department, the reason for the exit. (28) The raiding story implies that it is low unmeasured productivity, and not seniority, that results in lower pay. In contrast to senior faculty, who have low unmeasured productivity and do not generate outside interest, faculty who are so marketable that they move upward or laterally should have high unmeasured productivity. There is no reason their pay should be inversely related to seniority, even if productivity is not measured fully. If they are penalized for seniority, and especially, if they are penalized as severely as faculty who remain with the university, we can rule out raiding as the source of negative seniority profiles.

The initial specification of Table 5, without controls for research productivity, suggests that faculty who are bid away (move to better, or at least comparable, positions) are penalized for their seniority but that the penalty may be less severe. But, as previously noted, these movers are more productive than other faculty. When controls are included for research output (column 2), results indicate that the penalty for faculty who are bid away is as severe as for faculty overall. (29) Their pay may rise faster than that of other faculty but not commensurately with their greater productivity. These findings are inconsistent with raiding theory.

Next, we expand the model to allow for different seniority profiles for all classifications of movers (Table 5, column 3). Based on the differential intercepts, tenure denials are paid less than faculty who stay with the university. Apart from the lower intercept, tenure denials are penalized an additional 1.0% for each year of seniority relative to faculty who remain with the university. But these results may be biased by endogeneity of wages and mobility. If the department realizes that a faculty member is on the way out, whether because of tenure denial or some other reason, the department has an incentive to give the faculty member a lower pay increment the final year (to reallocate the funds to continuing faculty). In that event, keeping the last year's observation on pay in the sample could bias downward returns to seniority for movers.

One solution is to delete from the sample the final pay period for faculty leavers. When we adopt this approach (column 4), all three seniority interaction terms become more positive (less negative), and the coefficient of Seniority x Denial becomes insignificant, though it remains negative. Much of the wage penalty incurred by faculty who are denied tenure is apparently paid in their final year in the department. Once we remove this observation, there is no longer evidence that this group is penalized more for seniority than other groups of faculty. In fact, based on the estimates of column 4, all categories of movers receive seniority penalties comparable to those of nonmovers. (30) The key finding of this section is that the penalty to seniority persists even when we allow for different intercepts for faculty mobility status (to permit higher unmeasured productivity of faculty who are bid away and lower productivity of faculty who are denied tenure) and even when we allow for differential seniority penalties.

C. Rewards for a PhD from a Top 10 Program?

As argued previously, faculty with degrees from top 10 programs may experience a lower seniority penalty than other faculty for either of two reasons: they may have higher unmeasured productivity or universities may view them as more marketable and hence exert less monopsony power against these faculty. Sixty-four of the faculty in our sample (27%) graduated from a top 10 university. (31) Table 6 allows for differential returns to seniority for such faculty. Estimates indicate that these faculty, while still penalized for seniority, suffer a lower penalty than other faculty. Comparing columns 1-2, we see that the seniority advantage (lower penalty) for top 10 faculty shrinks once we control for productivity, consistent with their slight productivity advantage over other faculty. Based on the estimates of column 2, the penalty to seniority is lower for top 10 faculty than for other faculty (8.1% after 10 yr rather than 10.5%).

The remaining columns of Table 6 present the most comprehensive specification of the model, first for the full sample (column 3) and then after dropping final-year observations for faculty who leave (column 4). As in Table 5, once we drop observations for faculty leavers, seniority penalties do not differ significantly by mobility status. According to the estimates of column 4, the overall seniority penalty is approximately 0.9% per year for faculty who did not graduate from top 10 programs (9.2% after 10 years). Faculty from top 10 programs are also penalized for seniority but less severely than other faculty (6.7% after 10 years). (32) For all categories of faculty studied, returns to seniority are negative, substantial, and statistically significant.

D. Allowing for Match Quality and Changes in the Reward Structure

Thanks to the information revolution (Internet, Web pages, posting of information by the American Economic Association, etc.), the academic job market operates more efficiently now than in the past. In this section, we look at whether returns to seniority and experience have changed in recent years, as might be expected because of greater availability of information and perhaps fewer bad matches (where the mix of faculty skills and attributes differs from that sought by the university).

According to the literature, good matches between workers and employers tend to last longer than poor matches and to be associated with higher pay. (33) Unless match quality is accounted for, the resulting correlation between seniority and the error term generates a positive bias in the estimates of the returns to seniority. To allow for the effect of match quality, we present estimates based in the two-step procedure of Topel (1991) as well as random-effects estimates. We provide estimates for two subperiods, 1975-1994, which can be thought of as the pre-Internet period, and 1995-2004 (post-Internet). (34) In both cases, we drop the last observation for movers and calculate cumulative returns to seniority and experience at 15 yr, a value not far from the mean for the sample (mean seniority is 14 yr, and mean experience is 16 yr). Results appear in Table 7.

There is no evidence that returns to experience have changed over time. Regardless of estimation technique and time period, cumulative returns to 15 years of experience are roughly 30%. In contrast, random-effects and Topel estimators tell different stories about returns to seniority. According to random-effects estimates (columns 3 and 5), the seniority penalty has increased over time. But Topel estimates for the two periods are comparable (columns 4 and 6), with no signs that the penalty has increased. Instead, these estimates indicate that the role of match quality bias has eroded over time.

Apparently, the correlation between (unobserved) match quality and seniority was important in the early period, causing random-effects estimates of the seniority penalty to be biased toward zero, as Bratsberg, Ragan, and Warren (2003) found in their analysis of data from 1975 to 1995. (35) In the latter period, however, random-effects and Topel estimates are not appreciably different, suggesting that traditional estimation techniques are no longer biased. By implication, recent hires are better matched with the universities than faculty hired in the earlier era. This finding is consistent with both sellers and buyers being better informed than in the past, perhaps as a consequence of the Internet revolution, so that fewer truly bad matches occur.

Finally, the change in the market for economists has mixed implications for mobility. The better quality of matches might reduce turnover, but greater information makes it easier to move, which is an offsetting effect. What we can conclude is that the nature of mobility has changed. The share of mobility caused by poor matches is lower now than in the past, and that is why match quality bias has eroded.

VI. SUMMARY

This study uses information on faculty mobility to address the issue of why returns to faculty seniority are negative. The two hypotheses that have been advanced are monopsony power by universities and faculty raiding. According to the raiding story, senior faculty have low unmeasured productivity, which creates a negative bias in estimated returns to seniority. The raiding story has several implications: (1) the bias should dissipate as better measures of research output are added to the wage equation; (2) the estimated seniority penalty should be lower if we control for faculty who are bid away--and presumably have high unmeasured productivity; and (3) estimated returns to seniority should not be negative for faculty who are bid away. To examine these propositions, we collected the most detailed measures of productivity used to date and we allowed the wage structure to differ for faculty who move upward or laterally (presumably the targets of raids) and for other categories of faculty.

We followed all tenure-track faculty at five universities over a 30-yr time period. As such, our analysis provides a detailed look at one segment of the academic labor market: midlevel, doctoral granting departments of economics in the Midwest. In our sample, about two-thirds of faculty who leave for alternative employment move downward; less than one-quarter moves upward. Faculty turnover is heaviest in the years immediately preceding the tenure decision.

There is a substantial penalty for seniority with the university. The penalty moderates when controls are added for research productivity, pointing to an inverse relationship between seniority and productivity--yet, in all specifications of the model, returns to seniority are negative and significant. According to the random-effects estimates of Table 7, after accounting for productivity, the penalty for faculty overall is 8.9% at 10 yr of seniority and 12.2% at 15 yr. When we switch to a Topel estimator, the estimated penalty rises modestly (to 14.6% at 15 yr). Loyalty may be rewarded in some quarters, but not in academia.

The seniority penalty is little changed when we allow for differential intercepts and differential returns to seniority by mobility status. Faculty who end up moving to better or comparable positions were penalized as severely for seniority while they were in our sample as faculty who stay. These results are incompatible with the raiding hypothesis.

Even faculty from top 10 universities receive negative returns to seniority, though their penalty is less than that of faculty who graduated from lesser ranked institutions. The better treatment of top 10 faculty could potentially reflect higher unmeasured productivity, but it is also consistent with the model of monopsony if universities view top 10 faculty as more likely to receive outside offers or to respond to outside offers.

Faculty are rewarded highly for research productivity. According to the baseline version of the model, returns to a sole-authored, 10-page article are estimated to be 3.9% if published in a top 10 journal, 1.9% if in a tier 2 journal, 1.5% if in a tier 3 journal, and 0.9% if in an unranked journal. Books also increase faculty salary. The payoff appears to be greatest for the first edition of a textbook, perhaps because of the national visibility a textbook, can create. Subsequent editions of the text, research books, edited books, and book chapters increase pay by 1%-2% compared to the 4% return for a new textbook. Publications that are cited generate additional compensation.

The lower pay of senior faculty cannot be attributed to differences in impending research flows or a tendency to have dated publications whose value has depreciated over time. Accounting for prospective articles and for depreciation in returns to past publications, another innovation of this study, does not appreciably alter any of the other results, including the findings on seniority. Research is rewarded even prior to publication, but returns are considerably greater once the articles are in print. For articles in ranked journals, returns are slightly higher for papers published in the past 5 years.

When research output is excluded from the analysis, faculty who serve as major professor appear to be compensated for this activity, but their higher pay is actually the consequence of their research output. Students are drawn to faculty with strong research reputations, but advising is not directly rewarded.

When we allow for a change in pay structures in the pre-Internet and post-Internet periods, random-effects estimates suggest that the seniority penalty has increased, but Topel estimates, which account for match quality, indicate otherwise. The former period was characterized by correlation of match quality and seniority, which caused random-effects estimates to understate the seniority penalty. In recent years, there have been fewer poor matches, presumably the consequence of greater availability of information. Now that match quality bias is no longer important, traditional estimation techniques no longer understate the penalty to seniority.

In conclusion, the results of this study help us understand better the operation of academic labor markets by providing insights on faculty mobility and pay structures. Most important, they lead us to reject the raiding interpretation of the negative returns to faculty seniority.

REFERENCES

Abowd, J., F. Kramarz, and S. Roux. "Wages, Mobility and Firm Performance: Advantages and Insights from using Matched Worker-Firm Data." Economic Journal, 116, 2006, F245-85.

Abraham, K. G., and H. A. Farber. "Job Duration, Seniority, and Earnings." American Economic Review, 77, 1987, 278-97.

Altonji, J. G., and R. A. Shakotko. "Do Wages Rise with Seniority?" Review of Economic Studies, 54, 1987, 437-59.

Altonji, J. G., and N. Williams. "Do Wages Rise with Seniority? A Reassessment." Industrial and Labor Relations Review, 58, 2005, 370-97.

Ault, D. E., L. G. Rutman, and T. Stevenson. "Mobility in the Labor Market for Academic Economists." American Economic Review, 69, 1979, 131-36.

--. "Some Factors Affecting Mobility in the Labor Market for Academic Economists." Economic Inquiry, 20, 1982, 104-32.

Axarloglou, K., and V. Theoharakis. "Diversity in Economics: An Analysis of Journal Quality Perceptions." Journal of the European Economic Association, 1, 2003, 1402-23.

Barbezat, D. A. "The Effect of Collective Bargaining on Salaries in Higher Education." Industrial and Labor Relations Review, 42, 1989, 443-55.

Barbezat, D. A., and J. W. Hughes. "The Effect of Job Mobility on Academic Salaries." Contemporary Economic Policy, 19, 2001, 409-23.

Boal, W., and M. Ransom. "Monopsony in the Labor Market." Journal of Economic Literature, 35, 1997, 86-112.

Bratsberg, B., J. F. Ragan Jr., and J. T. Warren. "Negative Returns to Seniority: New Evidence in Academic Markets." Industrial and Labor Relations Review, 56, 2003, 306-23.

Brown, B. W., and S. A. Woodbury. "Seniority, External Labor Markets, and Faculty Pay." Quarterly Review of Economics and Finance, 38, 1998, 771-98.

Burton, M. P., and E. Phimister. "Core Journals: A Reappraisal of the Diamond List." Economic Journal, 105, 1995, 361-73.

Ciolek, M. T., and S. Whitfield. "The Past and Futures of Scholarly Internet." Rzeczpospolita, 2000. Accessed April 5, 2006. http://www.ciolek.com/PAPERS/ e-scholarship2000.html.

Coupe, T. "Revealed Performances: Worldwide Rankings of Economists and Economics Departments, 1990-2000." Journal of the European Economic Association, 1, 2003, 1309-45.

Coupe, T., V. Smeets, and F. Warzynski. "Incentives, Sorting and Productivity Along the Career: Evidence from a Sample of Top Economists." Journal of Law, Economics, & Organization, 22, 2005, 137-67.

Dustmann, C., and C. Meghir. "Wages, Experience, and Mobility" Review of Economic Studies, 72, 2005, 77-108.

Ehrenberg, R. G. "Adam Smith Goes to College: An Economist Becomes an Academic Administrator." Journal of Economic Perspectives, 13(1), 1999, 99-116.

Ehrenberg, R. G., H. Kasper, and D. I. Rees. "Faculty Turnover at American Colleges and Universities: Analyses of AAUP Data." Economics of Education Review, 10, 1991, 99-110.

Enomoto, C. E., and S. N. Ghosh. "A Stratified Approach to the Ranking of Economics Journals." Studies in Economic Analysis, 14(2), 1993, 74-93.

Ginther, D., and S. Kahn. "Women in Economics: Moving Up or Falling Off the Academic Ladder?" Journal of Economic Perspectives, 18(3), 2004, 193-214.

Gordon, N., T. Morton, and I. Braden. "Faculty Salaries: Is There Discrimination by Sex, Race, and Discipline?" American Economic Review, 64, 1974, 419-27.

Graves, P., J. Marchand, and R. Thompson. "Economics Departmental Rankings: Research Incentives, Constraints, and Efficiency." American Economic Review, 72, 1982, 1131-41.

Hamermesh, D. S., G. E. Johnson, and B. A. Weisbrod. "Scholarship, Citations and Salaries: Economic Rewards in Economics." Southern Economic Journal, 49, 1982, 472-81.

Harris, M., and B. Holmstrom. "Theory of Wage Dynamics." Review of Economic Studies, 49, 1982, 315-33.

Hirsch, B., R. Austin, J. Brooks, and B. Moore. "Economics Departmental Rankings: Comment." American Economic Review, 74, 1984, 822-26.

Kalaitzidakis, P., T. P. Mamuneas, and T. Stengos. "Rankings of Economics Journals." Journal of the European Economic Association, 1, 2003, 1346-66.

Katz, D. A. "Faculty Salaries, Promotions, and Productivity at a Large University." American Economic Review, 63, 1973, 469-77.

Lazear, E. P. "Raids and Offer Matching, in Research in Labor Economics, Vol. 8, edited by R. G. Ehrenberg Greenwich, CT: JAI Press, 1986, 141-65.

Monks, J., and M. Robinson. "The Returns to Seniority in Academic Labor Markets." Journal of Labor Research, 22, 2001, 415-27.

Moore, W. J., R. J. Newman, J. Raisian, and R. W. Thomas. "A Quality Adjustment Model of the Academic Labor Market: The Case of Economists." Economic Inquiry, 21, 1983, 241-54.

Moore, W. J., R. J. Newman, and D. Terrell. "Academic Pay in the United Kingdom and the United States: The Differential Returns to Productivity and the Lifetime Earnings Gap." Southern Economic Journal, 73, 2007, 717-32.

Moore, W. J., R. J. Newman, and G. K. Turnbull. "Do Faculty Salaries Decline with Seniority?" Journal of Labor Economics, 16, 1998, 352-66.

Ragan, J. F. Jr., and Q. N. Rehman. "Earnings Profiles of Department Heads: Comparing Cross-Section and Panel Models." Industrial and Labor Relations Review, 49, 1996, 256-72.

Ragan, J. F. Jr., J. T. Warren, and B. Bratsberg. "How Similar Are Pay Structures in 'Similar' Departments of Economics?" Economics of Education Review, 18, 1999, 347-60.

Ransom, M. R. "Seniority and Monopsony in the Academic Labor Market." American Economic Review, 83, 1993, 221-33.

Rees, D. I. "Does Unionization Increase Faculty Retention?" Industrial Relations, 33, 1994, 297-321.

Scott, L. C., and P. M. Mitias. "Trends in Rankings of Economics Departments in the U.S.: An Update." Economic Inquiry, 34, 1996, 378-400.

Siegfried, J. J., and W. Stock. "The Labor Market for New Ph.D. Economists." Journal of Economic Perspectives, 13(3), 1999, 115-34.

--. "The Market for New Ph.D. Economists in 2002." American Economic Review, 94, 2004, 272-85.

Tolbert, P. S., T. Simons, A. Andrews, and J. Rhee. "The Effect of Gender Composition in Academic Depart ments on Faculty Turnover." Industrial and Labor Relations Review, 48, 1995, 562-79.

Topel, R. "Specific Capital, Mobility, and Wages: Wages Rise with Job Seniority." Journal of Political Economy, 99, 1991, 145-76.

Tremblay, C. H., V. J. Tremblay, and B. Lee. "Field Publishing Performance of U.S. Economics Departments." Atlantic Economic Journal, 18, 1990, 37-48.

Tschirhart, J. "Ranking Economics Department in Areas of Expertise." Journal of Eeonomic Education, 20(2), 1989, 199-222.

Zakon, R. H. "Hobbes' Internet Timeline." Zakon Group LLC. 2005. Accessed April 5, 2006. http://www. zakon.org/robert/internet/timeline/.

BERNT BRATSBERG, JAMES F. RAGAN JR, and JOHN T. WARREN *

* The authors have benefited from the comments of Richard Edwards, Ronald Ehrenberg, Dan Hamermesh, Derek Neal, Oddbjorn Raaum, Kjetil Storesletten, Laura Ragan, Carol Tremblay, Vic Tremblay; seminar participants at Claremont Graduate University, Oregon State University, Kansas State University, the University of Nebraska, the University of Puerto Rico-Bayamon, the Frisch Centre, the Missouri Economics Conference; the co-editor: and two referees. We thank John Anderson, Dave Mandy, Craig MacPhee, Peter Mueser, Peter Orazem, Mike Podgursky, Jim Schmidt, and Joe Sicilian for assisting us in the collection of certain data. Part of the project was completed while Ragan was a visiting scholar at the Frisch Centre; he is grateful for the Centre's support.

Bratsberg: Senior Research Fellow, Ragnar Frisch Centre for Economic Research, Gaustadalleen 21, 0349 Oslo, Norway. Phone +47 2295 8818, Fax +47 2295 8825, E-mail bernt.bratsberg@frisch.uio.no

Ragan, Jr.: Professor, Department of Economics, Kansas State University, Waters Hall, Manhattan, KS 66506-4001. Phone 785-532-4582, Fax 785-532-6919, E-mail jfrjr@ksu.edu

Warren: Adjunct Professor, Department of Economics, Kansas State University, Waters Hall, Manhattan, KS 66506-4001. Phone 785-532-7357, Fax 785-532-6919, E-mail jomega@ksu.edu

doi: 10.1111/j.1465-7295.2009.00220.x

(1.) Related research shows that faculty turnover varies with gender composition of the department (Tolbert et al. 1995).

(2.) See Bratsberg, Ragan, and Warren (2003) for a review of this research. A more recent study is Moore, Newman, and Terrell (2007).

(3.) If one can come up with a third explanation for why returns to seniority are negative (other than raiding and monopsony power) then rejection of the raiding hypothesis would not imply acceptance of the monopsony model, but we are unaware of a compelling alternative to these two models. As one referee observed, administrative constraints on pay increases of existing faculty coupled with growing demand for PhD economists may have reduced the relative pay of senior faculty. But we interpret this story in terms of the monopsony model: faculty must move to realize their market salary; universities reap rents from faculty who are tied down.

(4.) For example, Ransom (1993) found a greater seniority penalty for faculty from "high-quality universities" than for faculty in general.

(5.) This finding is at odds with the positive returns to seniority often found for other professions (e.g., see Ransom 1993). One explanation may be the tenure system in academia. Once a faculty member is tenured, the list of available alternative jobs shrinks considerably, limiting mobility and increasing potential monoposony power. Moreover, the incentive effects of seniority-based pay, found in other sectors of the economy, are based on the threat of dismissal, which effectively disappears with academic tenure. Finally, in nonacademic professions, an employee who leaves may take along clients with whom the employee has cultivated a professional relationship. To deter such exits, firms in the financial sector, law, accounting, and so forth have an incentive to reward seniority.

(6.) The exceptions are Katz (1973), Hamermesh, Johnson, and Weisbrod (1982), and Ragan, Warren, and Bratsberg (1999). These studies show that research quality is a more important determinant of pay than quantity of research.

(7.) The coefficient is -0.0064 (se = 0.0026) or -0.0082 (se = 0.0023), depending on the specification.

(8.) Among U.S. universities, the departments in our sample are ranked between 42 and 102 by Tschirhart (1989) and between 51 and 101 by Coupe (2003). Four of the departments appear between 69 and 86 in the ranking of the top 88 U.S. universities by Kalaitzidakis, Mamuneas, and Stengos (2003). Tremblay, Tremblay, and Lee (1990) place four of the universities in the 51 92 range.

(9.) Although 1975/1976 is the initial year of the sample (the first year for which we have pay data), we include faculty whose employment at the university began prior to 1975/1976.

(10.) Data on faculty mobility were typically obtained from vitae posted on the Web. When vitae were missing, we searched the American Economic Association directory, googled the faculty name, or called the person's former department to solicit information. Through the same sources and university Web pages, we determined the year in which a person received the doctoral degree and the school conferring it.

(11.) The time period and source of the rankings are 1974-1978 (Graves, Marchand, and Thompson 1982), 1978-1983 (Hirsch et al. 1984), 1984-1993 (Scott and Mitias 1996), and 1990-2000 (Coupe 2003).

(12.) In all cases, classifications were based on a thoughtful a priori discussion of the move, but there are so few moves in our data set whose direction is debatable that we are confident that empirical estimates are not sensitive to the classification system adopted.

(13.) These data are consistent with the findings of Ault, Rutman, and Stevenson (1979, 1982) that the average academic move is downward.

(14.) Although "tenure denial" is defined in terms of seniority of 4-7 years, empirical results are not sensitive to this classification. If tenure denial is alternatively defined as downward mobility of faculty with experience of 4-7 years, results are similar. Likewise, our conclusions are unchanged when we define tenure denial more broadly as downward mobility for faculty with seven or fewer years of seniority.

(15.) For individual faculty, changes in seniority and experience are perfectly correlated, which precludes estimation of [[mu].sub.i] as a faculty fixed effect. An advantage of random effects over ordinary least squares is that it allows for individual heterogeneity, and Breusch-Pagan tests overwhelmingly favor the random-effects specification over ordinary least squares for our sample.

(16.) In supplemental regressions, we add the term [[eta].sub.ij] to account for match quality of faculty member i at university j and use the Topel (1991) two-step estimator.

(17.) EconLit covers the period 1969 to present. Earlier data were obtained from the American Economic Association, Index of Economic Articles. Our measure of articles excludes book reviews, papers in conference proceedings, and articles in journals not covered by EconLit.

(18.) These journals are American Economic Review, Econometrica, Economic Journal, Economica, International Economic Review, Journal of Economic Theory, Journal of Political Economy, Quarterly Journal of Economics, Review of Economic Studies, and Review of Economics and Statistics.

(19.) Axarloglou and Theoharakis (2003) and Kalaitzidakis, Mamuneas, and Stengos (2003) rank Journal of Economic Perspectives and European Economic Review among the top 15 journals. Burton and Phimister (1995) have Economic History Review number 10 based on their application of data envelope analysis to citation data.

(20.) Data on books were obtained from EconLit from 1988 to present and from WorldCat for earlier years (prior to EconLit coverage). Only original book chapters were included, not reprints.

(21.) An exception is Ragan and Rehman (1996) who estimate returns to textbooks and research books across faculty from multiple departments at one university. They report larger returns for research books than textbooks.

(22.) We thank Derek Neal for pointing out that if past publications are heavily discounted and prospective publications rewarded, misspecification of the reward structures could potentially explain the negative returns to seniority found in the literature. By allowing for lower returns to older publications and positive returns to prospective publications, we are able to test for this possibility.

(23.) Mobility patterns may have shifted favorably for females over the sample period. During the period 1990-2004, six females were "bid away" and four were denied tenure. In the period before 1990, only one female was bid away (in 1989), while four were denied tenure. For males, the patterns of being bid away and being denied tenure did not change between subperiods.

(24.) Based on the estimates of column 2, the penalty for 10 yr of seniority is exp(-0.01392 x 10 + 0.0000288 x 100) 1 = exp(-0.1363) 1 = 0.1274. The negative returns to seniority are not driven by a subset of our sample. When we allow returns to seniority to differ by university, results point to significant and negative returns at each university.

(25.) One scholar confided to one of the authors that his research book sold only 50 copies. More generally, textbooks sell at higher levels than research books. The relative rewards to the different categories of books may also depend on the relative importance of teaching at the institution, in which case the returns to textbooks are likely to be lower at more elite institutions.

(26.) The 0.6 premium is the sum of the coefficients of Recent Pages Cat1/10 and Recent Articles (0.0090-0.0031).

(27.) When we allow for two sets of lags, we find that the return to articles is lowest for articles published more than 10 yr ago, but articles published 6-10 yr ago are not discounted relative to those published in the past 5 yr. One reason may be that the effect of any depreciation in the value of articles published in the 6- to 10-yr timeframe is offset by the fact that the returns to recent articles are not fully received until years after the publication (because of budget pressures in any given year) or because of departmental policy to explicitly value past publications (e.g., by rewarding research reputation).

(28.) Although we cannot identify which faculty remain with the university after receiving an outside offer and then a counteroffer, counteroffers soften the seniority penalty. If we excluded faculty who receive counteroffers, the seniority penalty of nonmovers would be even greater. Because those with positive pay shocks are less likely to move, we may underestimate the seniority penalty for nonmovers. Ideally, we would like to allow for endogeneity of faculty mobility, both upward and downward, but that requires instrumental variables that affect mobility but not pay. Although we have been unable to come up with plausible instruments, another way to address endogeneity is to delete observations in the final year or two that movers are in the department, when low pay could plausibly result from knowledge by the department chair that such faculty are leaving. This approach is adopted in the following section.

(29.) If we alternatively define mobility in terms of upward, lateral, and downward moves (rather than bid away, tenure denials, and other moves), we find that faculty who move to better universities are penalized at least as severely as faculty overall, which is at odds with the raiding hypothesis.

(30.) Results are similar if we drop the final two observations for movers.

(31.) The heavy concentration of faculty from top 10 universities is consistent with the finding of Moore et al. (1983) that "most academic appointments follow a 'downstream' pattern."

(32.) As one might expect, mobility patterns differ by pedigree of PhD degree: Faculty in our sample are more likely to be bid away and less likely to be denied tenure if they graduated from a top 10 program. Nonetheless, when we relax the specification of Table 6 to allow a different pattern of coefficients for top 10 and other faculty, our conclusions still hold. The finding that top 10 faculty who are bid away are penalized as severely as faculty who remain is especially difficult to reconcile with the raiding interpretation of the observed negative seniority profiles.

(33.) See, for example, Abraham and Farber (1987), Altonji and Shakotko (1987), Altonji and Williams (2005), and Topel (1991). Recent studies point to substantial heterogeneity in returns to seniority across different segments of the labor market (Abowd, Kramarz, and Roux 2006; Dustmann and Meghir 2005).

(34.) The Internet had become an important tool by 1995. Zakon (2005) reports that 1995 was characterized by WWW, search engines, and the introduction of JAVA. That year, the NSF linked supercomputing centers to create "the very high speed Backbone Network Service" and the "main U.S. traffic [was] now routed through interconnected network providers." The Internet had become very important on university campuses, with edu second only to corn in the ranking of domains by host numbers. Ciolek and Whitfield (2000) concur, noting that "In the mid-1990s as it finally entered most scholarly institutions and became genuinely accessible to all, some scholars immediately mounted home pages, published electronic articles and sought funds for online projects."

(35.) Bratsberg, Ragan, and Warren (2003) report that the penalty to 15 yr of seniority was 8.0% when estimated with the random-effects model and 14.6% when estimated with the Topel model. Based on our more elaborate specification and an additional 9 yr of data, our estimates are 12.2% and 14.6%, respectively.

TABLE 1
Faculty Status after Leaving the Sample

Status Frequency %

Retirement 49 32.0
Death/illness 4 2.6
Business 6 3.9
Federal government 10 6.6
State government 6 3.9
Foreign university 16 10.5
Foreign other 4 2.6
Other U.S. university 51 33.3
Other unit in university 7 4.6
Total 153 100.0

TABLE 2
Sample Descriptive Statistics

Variable Mean SD Minimum Maximum

In (salary) 10.568 0.248 9.923 11.365
Move 0.232 0.422 0 1
Bid Away 0.082 0.274 0 1
Denial 0.062 0.242 0 1
Other Move 0.088 0.283 0 1
Top 10 0.249 0.432 0 1
Seniority 13.900 10.358 0 42
Experience 16.001 10.335 0 42
Female 0.072 0.258 0 1
Chair 0.049 0.217 0 1
Years Chair 0.527 1.755 0 15
Past Chair 0.069 0.253 0 1
Past Dean 0.003 0.051 0 1
Advisees 2.134 3.763 0 41
Articles 7.307 8.488 0 93
Pages Cat l 8.578 14.977 0 106.506
Pages Cat2 13.181 22.282 0 181.288
Pages Cat3 11.237 28.313 0 285.063
Recent Articles 2.312 3.154 0 23
Recent Pages Catl 1.975 5.360 0 55.493
Recent Pages Cat2 4.088 9.636 0 98.522
Recent Pages Cat3 3.087 9.338 0 96.289
Prospective Articles 1.670 2.508 0 39
Prospective Pages Catl 1.078 3.370 0 31.059
Prospective Pages Cat2 2.764 6.871 0 52.616
Prospective Pages Cat3 2.176 6.542 0 69.771
Book Text 0.181 0.519 0 3
Text Revised 0.145 0.760 0 9
Book Research 0.269 0.852 0 10
Book Edited 0.182 0.829 0 13
Book Chapter 0.493 1.529 0 22
Citations 33.285 61.705 0 837

Notes: Sample consists of 2,698 observations of 238 faculty.

TABLE 3
Sample Means by Mobility Status

Variable Nonmovers Bid Away Denial Other Move

In (salary) 10.605 10.526 10.271 10.494
Top 10 0.269 0.253 0.173 0.118
Seniority 16.218 6.167 2.393 9.004
Experience 18.403 8.538 3.226 11.013
Female 0.047 0.217 0.190 0.063
Chair 0.056 0.068 0 0.013
Years Chair 0.644 0.276 0 0.110
Past Chair 0.083 0 0 0.063
Past Dean 0.003 0 0 0
Advisees 2.534 0.432 0.060 1.696
Articles 7.794 8.584 1.161 6.207
Pages Cat] 9.331 7.311 0.993 8.554
Pages Cat2 14.095 13.791 1.352 13.013
Pages Cat3 12.546 11.210 0.526 7.413
Recent Articles 2.224 3.819 0.958 2.633
Recent Pages Cat] 1.834 3.414 0.504 2.911
Recent Pages Cat2 3.845 8.373 1.233 4.238
Recent Pages Cat3 3.273 3.728 0.498 2.695
Prospective Articles 1.466 3.434 1.060 2.241
Prospective Pages Catl 0.894 2.391 0.742 1.700
Prospective Pages Cat2 2.219 8.189 2.335 2.772
Prospective Pages Cat3 2.224 3.134 1.285 1.492
Book Text 0.217 0.113 0 0.051
Text Revised 0.186 0.018 0 0
Book Research 0.285 0.443 0.030 0.139
Book Edited 0.204 0.204 0 0.097
Book Chapter 0.577 0.425 0.036 0.139
Citations 36.681 41.489 1.857 18.224
Observations 2,072 221 168 237
Faculty 138 34 35 31

TABLE 4
Random-Effects Estimates for Faculty Pay Structures

Independent Variable (1) (2)

Seniority -.0139 *** (.0023) -.0139 *** (.0021)
[Seniority.sup.2]/100 .0104 ** (.0050) .0029 (.0047)
Experience .0392 *** (.0025) .0365 *** (.0022)
Experience 2/100 -.0543 *** (.0051) -.0485 *** (.0047)
Female .0124 (.0274) .0126 (.0245)
Chair .0862 *** (.0113)
Years Chair .0100 *** (.0021)
Past Chair .0456 *** (.0140)
Past Dean .2837 *** (.0386)
Advisees .0107 *** (.0008)
Articles
Recent Articles
Prospective Articles
(Pages Catl)/10
(Recent Pages Catl)/10
(Prospective Pages Catl)/10
(Pages Cat2)/10
(Recent Pages Cat2)/10
(Prospective Pages Cat2)/10
(Pages Cat3)/10
(Recent Pages Cat3)/10
(Prospective Pages Cat3)/10
Book Text
Text Revised
Book Research
Book Edited
Book Chapter
Citations/100
[Citations.sup.2]/10,000

Independent Variable (3) (4)

Seniority -.0120 *** (.0018) -.0113 *** (.0018)
[Seniority.sup.2]/100 .0169 *** (.0040) .0157 *** (.0041)
Experience .0234 *** (.0019) .0231 *** (.0019)
Experience 2/100 -.0416 *** (.0041) -.0412 *** (.0041)
Female .0376 * (.0207) .0417 ** (.0203)
Chair .0765 *** (.0092) .0801 *** (.0092)
Years Chair .0164 *** (.0017) .0163 *** (.0017)
Past Chair .0305 *** (.0117) .0311 *** (.0118)
Past Dean .3925 *** (.0317) .3892 *** (.0316)
Advisees -.0003 (.0008) -.0001 (.0008)
Articles .0085 *** (.0007) .0093 *** (.0007)
Recent Articles -.0031 ** (.0012)
Prospective Articles .0005 (.0010)
(Pages Catl)/10 .0295 *** (.0032) .0308 *** (.0036)
(Recent Pages Catl)/10 .0090 ** (.0038)
(Prospective Pages Catl)/10 .0156 *** (.0057)
(Pages Cat2)/10 .0104 *** (.0018) .0083 *** (.0022)
(Recent Pages Cat2)/10 .0079 *** (.0028)
(Prospective Pages Cat2)/10 .0040 (.0028)
(Pages Cat3)/10 .0060 *** (.0013) .0046 *** (.0015)
(Recent Pages Cat3)/10 .0039 (.0028)
(Prospective Pages Cat3)/10 .0065 ** (.0033)
Book Text .0418 *** (.0077) .0413 *** (.0077)
Text Revised .0198 *** (.0045) .0195 *** (.0046)
Book Research .0201 *** (.0037) .0196 *** (.0037)
Book Edited .0129 ** (.0051) .0099 * (.0052)
Book Chapter .0147 *** (.0022) .0150 *** (.0023)
Citations/100 .0406 *** (.0118) .0462 *** (.0125)
[Citations.sup.2]/10,000 -.0141 *** (.0014) -.0149 *** (.0014)

Notes: Dependent variable is the natural logarithm of annual pay.
The regressions contain dichotomous variables to control for
university and academic year. The sample consists of 2,698
observations of 238 faculty.

* Statistically significant at the .10 level; ** statistically
significant at the .05 level; *** statistically significant at
the .01 level (two-tailed test).

TABLE 5
Differential Returns to Seniority for Movers

Independent Variable (1) (2)

Bid Away .0145 (.0239) .0013 (.0201)
Denial
Other Move
Seniority x Bid Away .0033 * (.0018) -.0020 (.0016)
Seniority x Denial
Seniority x Other Move
Seniority -.0142 *** (.0021) -.0110 *** (.0018)
[Seniority.sup.2]/100 .0035 (.0047) .0152 *** (.0041)
Control for Research? No Yes
Sample Full Full

Independent Variable (3) (4)

Bid Away -.0237 (.0204) -.0293 (.0201)
Denial -.0707 *** (.0212) -.0757 *** (.0211)
Other Move -.0158 (.0208) -.0234 (.0206)
Seniority x Bid Away -.0018 (.0016) .0001 (.0017)
Seniority x Denial -.0106 *** (.0035) -.0059 (.0046)
Seniority x Other Move -.0019 * (.0010) -.0002 (.0011)
Seniority -.0092 *** (.0018) -.0093 *** (.0018)
[Seniority.sup.2]/100 .0136 *** (.0041) .0119 *** (.0041)
Control for Research? Yes Yes
Sample Full Drops last
 observation for movers

Notes: Dependent variable is the natural logarithm of annual
salary. Regressions control for experience, gender,
administrative experience, advising, university, year, and
faculty random effects. The estimates of columns 2-3 include all
research productivity variables (those listed in Table 4, column
4). The full sample used for columns 1-3 consists of 2,698
observations of 238 faculty, and the restricted sample for column
4 of 2,568 observations of 238 faculty.

* Statistically significant at the .10 level; *** statistically
significant at the .01 level.

TABLE 6
The Effect of a Top 10 Degree

Independent Variable (1) (2)

Top 10 -.0197 (.0197) -.0040 (.0165)
Bid Away
Denial
Other Move
Seniority x Top 10 .0042 *** (.0006) .0027 *** (.0006)
Seniority x Bid Away
Seniority x Denial
Seniority x Other Move
Seniority -.0161 *** (.0021) -.0129 *** (.0018)
[Seniority.sup.2]/100 .0073 (.0047) .0180 *** (.0041)
Control for Research? No Yes
Sample Full Full

Independent Variable (3) (4)

Top 10 -.0105 (.0156) -.0091 (.0152)
Bid Away -.0218 (.0203) -.0274 (.0199)
Denial -.0692 *** (.0211) -.0739 *** (0210)
Other Move -.0174 (.0207) -.0244 (.0205)
Seniority x Top 10 .0026 *** (.0006) .0027 *** (.0006)
Seniority x Bid Away -.0015 (.0016) .0005 (.0017)
Seniority x Denial -.0106 *** (.0035) -.0060 (.0046)
Seniority x Other Move -.0012 (.0010) .0006 (.0011)
Seniority -.0109 *** (.0018) -.0110 *** (.0018)
[Seniority.sup.2]/100 .0156 *** (.0041) .0139 *** (.0041)
Control for Research? Yes Yes
Sample Full Drops last
 observation for movers

Notes: Dependent variable is the natural logarithm of annual
salary. Regressions control for experience, gender,
administrative experience, advising, university, year, and
faculty random effects. The estimates of columns 2-3 include all
research productivity variables. The full sample used for columns
1-3 consists of 2,698 observations of 238 faculty, and the
restricted sample for column 4 of 2,568 observations of 238
faculty.

* Statistically significant at the .10 level;; **statistically
significant at the .05 level; ***statistically significant at the
.01 level.

TABLE 7
Returns to Seniority and Experience by Subperiod--Comparing
Random-Effects and Topel Estimates

 Full Sample

 Random Effects Topel
Independent Variable (l) (2)

Seniority -.0107 *** (.0018) -.0151 *** (.0035)
Seniority 2/100 .0134 *** (.0041) .0308 *** (.0112)
Experience .0230 *** (.0019) .0267 *** (.0039)
Experience 2/100 -.0404 *** (.0041) -.0537 *** (.0113)
Seniority penalty -.1302 *** (.0225) -.1575 *** (.0354)
 at 15 yr
Cum returns .2527 *** (.0237) .2800 *** (.0385)
 to experience at 15 yr

 1975-1994

 Random Effects Topel
Independent Variable (3) (4)

Seniority -.0088 *** (.0024) -.0204 *** (.0053)
Seniority 2/100 .0243 *** (.0056) .0546 *** (.0160)
Experience .0253 *** (.0025) .0306 *** (.0056)
Experience 2/100 -.0513 *** (.0054) -.0734 *** (.0156)
Seniority penalty -.0767 *** (.0297) -.1828 *** (.0536)
 at 15 yr
Cum returns .2643 *** (.0305) .2933 *** (.0562)
 to experience at 15 yr

 1995-2004

 Random Effects Topel
Independent Variable (5) (6)

Seniority -.0139 *** (.0028) -.0083 ** (.0038)
Seniority 2/100 .0143 ** (.0072) -.0063 (.0131)
Experience .0226 *** (.0032) .0209 *** (.0045)
Experience 2/100 -.0374 *** (.0078) -.0179 (.0141)
Seniority penalty -.1756 *** (.0340) -.1390 *** (.0417)
 at 15 yr
Cum returns .2551 *** (.0379) .2731 *** (.0479)
 to experience at 15 yr

Notes: Regressions control for gender, research productivity,
university, and academic year (see specification reported in
Table 4, column 4), and sample drops the last observation for
movers. The sample in columns 1-2 consists of 2,598 observations
of 238 faculty, columns 3-4 have 1,771 observations of 177
faculty, and columns 5-6 have 827 observations of 142 faculty.
Standard errors in columns 2, 4, and 6 are adjusted for sampling
errors in first-step estimates.

 * statistically significant at the .01 level; ** statistically
significant at the .05 level; *** statistically significant at
the .01 level.

FIGURE 1

Direction of Move for Those Leaving the Sample

Downward (66%)
Laterally (11%)
Upward (23%)

Note. Data are limited to faculty who obtain alternative employment.

Note: Table made from pie chart.