The effect of gender-sorting on propensity to coauthor: implications for academic promotion.

McDowell, John M. ; Smith, Janet Kiholm

THE EFFECT OF GENDER-SORTING ON PROPENSITY TO COAUTHOR: IMPLICATIONS FOR ACADEMIC PROMOTION

A cohort sample of Ph.D. economists indicates a significant propensity for

researchers to select coauthors of the same sex. This gender-sorting contributes to lower

article production for women. Further, we find evidence of bias in academic promotion when

single-authored and coauthored articles carry the same weight in promotion and salary

decisions. The evidence explains, in part, why women academics wait longer for

promotion and are not as likely to be promoted as men. Among the effects of gender-sorting

is self-selection of women into larger departments where they are more likely to find

colleagues of the same sex.

I. INTRODUCTION

A number of empirical studies have documented lower salaries, research productivity and promotion rates for women faculty members than for men.(1) For example, Cole and Cole [1973] find that women scientists produce fewer papers and are also less likely to hold the ranks of Associate or Full Professor. Hansen et al. [1978] find that women economists produce fewer publications than men.(2) Johnson and Stafford [1974] and Farber [1977] present evidence that women economists have a lower chance of promotion. Weiss and Lillard [1982] find that women on average take almost twice as long as men to attain the rank of Associate or Full Professor, and attribute this slower advancement to lower productivity.(3)

In this paper we provide a partial explanation for productivity and promotion differences based on coauthorship patterns. We find evidence, consistent with Ferber and Teiman [1980], that economists tend to work with coauthors of the same sex. Given the proportions of men and women in the profession, this raises the possibility that women have fewer opportunities to collaborate, especially if employed by small departments.

While the reasons underlying gender-sorting in publishing are not known,(4) our evidence indicates that collaboration does not lead to statistically significant increases (or decreases) in productivity if coauthored articles are discounted by the number of authors, n. In spite of this, the effect of gender-sorting on academic rank is not neutral since there appears to be a distortion in promotion decisions introduced by the lack of weighting of coauthored work.

In empirical tests of the relationship of academic rank to research output we are unable to reject the hypothesis that single-authored and coauthored publications are weighted equally. Conversely, we do reject the hypothesis that the assigned weight is 1/n, the expected weight if departments seek to maximize total research output. Therefore, gender-sorting, which reduces the opportunities of women to coauthor, is one explanation for why women academics wait longer for promotion and are promoted at lower rates than men. Our results indicate that administrative decisions on how to weight coauthored articles may work to the detriment of women, especially in disciplines like economics where the proportions of men and women are unequal.

II. CHARACTERISTICS OF THE DATA

The dataset is a cohort sample of eighty-nine men and eighty-nine women receiving Ph.D.s in economics from the top twenty institutions between 1968 and 1975.(5) The sample period is necessarily truncated at 1975 so that records and career changes could be tracked for a full ten years following receipt of the Ph.D. Thus, publication data cover the period 1969-1986. Data were collected on experience, field of specialty, dates of promotion, academic affiliation, and number of articles published. Appendix A presents definitions of the variables and identifies the data sources. Appendix B contains descriptive sample statistics for the individuals in the study.

We determined each person's publication record by reference to the American Economic Association's Index of Economic Articles. For each article, we noted the journal name, the number of coauthors and their sex. Of the 464 articles produced by individuals with academic affiliations, 41 percent are coauthored (35 percent by two authors and 6 percent by three or more authors). While the percent of articles coauthored is not significantly different for women than men, our analysis indicates that this is because women tend to be employed by larger departments than men, where the probability of finding a suitable coauthor is higher.

Evidence of Gender-Sorting

In an article exploring the causes of lower research productivity for women, Ferber and Teiman [1980] reason that even if men constitute a larger fraction of all economists, female economists are not disadvantaged in collaborating if the formation of research teams is independent of gender. They find, however, that team formation is not random with respect to sex and conclude that economists are less likely to work with colleagues of the opposite sex than of the same sex.

Using our sample, we also test for the effect of gender on team formation. Table I shows the actual number of coauthors sorted by sex. Women produced seventy-seven coauthored articles with ninety-one collaborators. Of those ninety-one coauthors, thirty (33 percent) were female as compared to an expected number of sixteen (17 percent) if team formation were independent of gender. This contrasts sharply with sampled males where only eight (6 percent) collaborate with women as compared to an expected twenty-two (17 percent). The difference between observed and expected gender combinations producing coauthored work is highly significant.(6) This result is consistent with Ferber and Teiman's finding that researchers tend to work with coauthors of the same sex significantly more often than suggested by chance.

Whether female academics are disadvantaged by gender-sorting behavior will depend on whether research productivity is adversely affected by a lack of opportunities to coauthor and on how administrators measure the individual contributions of coauthors when making decisions on salaries and promotions.(7) These issues are explored below.

III. EMPIRICAL TESTS

The first hypothesis we consider is that gender (female = 1) has a significant effect on the formation of coauthorship teams. Since we find that the gender of the researcher does affect the opportunity to coauthor, we consider two related issues. The first is how coauthorship affects research productivity; the second is how coauthored work is evaluated or weighed in academic promotion.

Women in a predominantly male profession may be disadvantaged in several respects by barriers to coauthorship. First, less opportunity to coauthor may result in less human capital formation which, in turn, leads to lower research output. Second, even if coauthorship does not enhance productivity, less coauthorship may lead to a lower likelihood of promotion if departments "count lines" rather than measure individual contributions of coauthors.

The Coauthorship Decision

Many factors contribute to the formation of research teams--the presence of complementary skills held by coauthors, proximity to coauthors, compatible schedules of prospective coauthors, etc. Hence, the probability that an article is coauthored is expected to depend on both individual and job-specific factors. To test the hypothesis that gender has a significant effect on the probability of coauthoring, we examine the decision to coauthor as a function of gender and other variables expected to be important determinants of the decision.

A recent paper by Lindsay and Maloney [1988] provides a theoretical basis for hypothesizing relationships between gender, coauthorship and department size. As is well established, in labor markets with perfect mobility and sufficient numbers of unprejudiced employers, prejudice produces segregation but the wage gap is arbitraged away. If, however, segregation is costly and some men prefer to work with other men more than with women (or if women prefer to work with women) then firms will tend to evolve toward "all male" or "all female" employment structures, but the segregation will be incomplete. In industries where female participation in the labor force is small,(8) the Lindsay and Maloney model predicts women will select employment in work teams consisting of proportionately more women.(9)

Applying the Lindsay and Maloney model to the academic labor market suggests that if prejudice is present, then women will coauthor with other women with a greater likelihood than would be implied by chance (a result documented in Table I). Another implication is that female researchers will tend to work alone more frequently than males so that gender is negatively related to the probability of coauthoring where female = 1. Based on the Lindsay and Maloney analysis, researchers (who want to coauthor) are expected to react to any observed prejudice by seeking positions in departments employing others of the same sex (and nonprejudiced individuals). It is reasonable to expect that the larger the department the more likely a match will be found between women seeking female coauthors with complementary skills. Hence, we expect women to self-select into larger departments, and the larger the department, the more likely the female researcher will produce coauthored work. Thus, the expected sign on the gender-department size interacted variable is positive.

Following existing models of the coauthorship decision, we include experience and other control variables to capture characteristics of the author's field of specialty.(10) Based on previous research, more experience is expected to increase the probability of coauthoring. The greater the number of economists in an author's field, the higher the probability of locating a collaborator. More rapid field growth raises the cost of delays in completing and publishing research. As a consequence, the relative returns to collaborative activity are enhanced. For authors switching fields of specialization, we expect a greater probability of coauthoring since collaboration allows talents to be pooled so that a single researcher need not be familiar with every facet of the project. Also, the likelihood of finding a collaborator is higher in work environments with more potential coauthors, hence we expect a positive sign for the department size coefficient.

We also consider two additional interacted variables. Gender-sorting may create greater incentives for women to select to work in larger (more populated) fields relative to men, hence we expect a positive sign on a gender-field size interaction. Also, women's academic experience may differ from men's, perhaps due to career interruptions (see, for example, McDowell [1982]). We therefore include a gender-experience interacted variable such that if women experience more significant career interruptions than men, then the expected impact on coauthoring is negative.

Results. We use a logistic specification that transforms the coauthorship variable into the log of the odds ratio that an article is coauthored. In this analysis, an observation consists of an article produced by an individual who has an academic affiliation at the time of publication. Given that individuals are not observed publishing the same number of times, each observation is weighted by the individual's total number of publications produced over the ten year period. Maximum likelihood estimates are reported in Table II for two model specifications. The first is the basic model that includes the gender-department size interacted variable. The second adds the field size and the job experience interacted variables.(11)

For both specifications the coefficient of the gender variable is negative. The positive sign on the gender-department size variable indicates, as predicted, that it is female economists in smaller departments who experience an especially low probability of coauthoring.(12) To evaluate the net impact of gender on coauthoring, we take the partial derivative of model (2) with respect to gender and solve separately for the underlying male and female probabilities of coauthoring. Figure 1 plots these probabilities as a function of department size. The effects of other variables are evaluated at their means. This evidence indicates that the probability of coauthoring is much lower for women in small departments and not until department size reaches sixty-four members do the probabilities converge for men and women.

Other coefficients that were signed are consistent with expectations. One finding of interest concerns the impact of experience on the propensity to coauthor. Previous research indicates that coauthorship may have a life-cycle element as less-experienced economists tend to work alone.(13) While our evidence supports this finding for men, experience does not appear to affect the probability of coauthorship for women, since the parameter estimate on the gender-experience interaction is roughly equal and opposite in sign to that on experience. If McDowell and Melvin's interpretation is correct, that less-experienced (untenured) economists tend to work alone in order to build their reputations, then our evidence indicates this accumulation process is slower for female economists.

The results of models (1) and (2) provide support for Lindsay and Maloney's model of discrimination.(14) If discriminatory attitudes are present, the expectation is that female researchers will self-sort into environments (departments) where the possibility of collaborating with other women is highest. The data do show that the distribution of female economists is skewed toward larger departments. The mean department size of women in our sample is 54.4, which is significantly higher than 41.9 for men (Appendix B). This evidence along with gender-sorting behavior is consistent with the presence of discriminatory attitudes toward coauthoring with women.(15)

The Impact of Coauthorship on Research Productivity

We now consider how coauthorship affects research productivity and whether coauthorship patterns adversely affect development of female economists' human capital. Collaboration may augment human capital by exposure to more and varied research and may result in synergies (economies of scale) such that two individuals working together can produce more articles than if they worked separately. Synergies in coauthorship, however, may be offset by shirking. That is, two individuals working together may produce fewer articles than if they worked separately. This is particularly true if the incentive system does not discount coauthored research. Thus, even if potential output is increased by coauthorship, individuals may be less motivated to produce if all papers of equal quality count equally toward tenure regardless of the number of coauthors.

To test the impact of coauthorship on research productivity, we measure output in terms of a scale effect. The dependent variable is an individual's total article count at the end of ten years of experience, where each article is discounted by the number of authors. Thus, other things equal, a single-authored article counts as one unit of research production, and an article produced by n authors counts as 1/[n.sup.th] of a unit.(16)

The hypothesis is that the cumulative number of weighted articles is a function of the propensity to coauthor, as measured by the percentage of total publications that are coauthored. Hence, if enhanced productivity dominates, then the incremental impact of coauthoring will be positive; if shirking dominates, then the impact will be negative.

Given previous findings of lower research productivity for female economists, article production is also modelled as a function of gender. To capture the differential impact of academic versus nonacademic experience, we include a measure of accumulated academic experience and expect that academic experience enhances productivity. We also expect fewer total articles the higher the quality of the articles. An age variable is included because of the possibility that individuals who enter academic careers later in life may have lower propensities to publish. Finally, results from Table II suggest department size may affect article output through its effect on team formation. Also, to the extent department size is correlated with a greater research emphasis, one expects article output will be higher for individuals in such environments.

Results. Table III presents the results of a regression estimated using the subsample of individuals in our database who were employed in an academic institution for at least one year during their ten years of post-doctorate experience and who published at least one article. The evidence does not support the hypothesis that cumulative article production is increased by an individual's overall propensity to coauthor. In fact, the sign of the percent coauthored coefficient is negative, though nonsignificant.

The results also show a negative impact of gender on research productivity. It appears that, on average, women produce fewer publications even if adjustments are made for the number of coauthors.(17) The negative coefficient has several possible interpretations. The productivity differences may be the result of reduced incentives to publish given a lower probability of promotion.(18) They may result from women substituting away from publishing, possibly into administration or teaching, because of difficulties encountered locating coauthors. Lower output may also reflect difficulties women encounter dealing with editors and referees as suggested by Ferber and Teiman. Finally, lower output may be a result of longer or more frequent career interruptions for women.

One might expect that when faced with barriers to coauthorship women would substitute into single-authored research. However, the mean percent of coauthored articles is not significantly different for women than men.(19) It may be misleading, however, to interpret this as evidence that women are not disadvantaged by gender-sorting. We have documented that women select larger departments as employers possibly in an effort to mitigate the effects of gender-sorting. Further, we cannot observe the employment decisions or research output levels that would have resulted from equal opportunities to coauthor.

The Impact of Research Productivity on Academic Rank

The evidence above indicates that, given existing incentives, coauthoring does not detract from, or enhance, overall productivity when articles are discounted by the number of coauthors. We now evaluate whether promotion decisions made by academic institutions reflect this evidence when establishing weights for single-authored and coauthored articles. We examine the relative impact of coauthored versus single-authored work as variables explaining an individual's academic rank for each of ten years of postdoctorate experience (i.e., for Associate or Full Professor, rank=1). We expect that departments attempting to maximize research productivity will apply 1/n discounting for coauthored work (quality constant) when designing promotion criteria. Also, if women or other individuals do not have equal opportunities for coauthoring, then 1/n weighting tends to mitigate the effects of reduced opportunities.

Since an individual can only achieve rank in an academic setting, we employ a bivariate probit technique. We model the joint determination of whether an individual is in an academic setting in any given year of experience and, if so, whether the individual has achieved the rank of Associate or Full Professor. Thus, two models are actually estimated: one explaining achievement of rank and the other explaining the probability of being an academic. Only those individuals who select academia in a given year enter the sample for rank equation estimation, but since these individuals are nonrandomly selected from the sample of all individuals, the correlation between disturbances is used to produce consistent estimates of the relationships.(20)

We include in the rank equation data on research output, including the number of coauthored articles produced and the number of single-authored articles produced at the time of the observation. The institutional quality variable is included to control for the quality of the individual's academic institution. The variable is measured as the number of AER-equivalent pages published per time period by department faculty standardized by number of faculty. We expect achievement of Associate or Full rank to be more difficult in higher quality departments. Quality of publications is controlled for, and we expect rank to be positively affected by quality. Since rank depends on experience, we include and experience variable and expect a positive relationship. To control for differences in male versus female academic experience, we also include a gender-experience interacted variable.

The focus of the study is not on the direct relationship, if any, between rank and gender, but rather on the indirect impact of gender on rank through the interactions between gender, coauthorship and research output. We include gender as a variable to capture other unobservable characteristics of the rank decision that may affect the probability of a woman achieving Associate or Full rank, but do not sign this variable a priori.

Results. Results for the rank equation appear in Table IV. The estimate is made using the entire sample of individuals where each is observed ten times (once for each year of experience).(21) We use all individuals, even those who did not publish since they may, depending on their institution's promotion criteria, achieve the rank of Associate or Full Professor with no published articles.

The coefficients for the variables measuring institutional quality and publication quality have the predicted signs. It does not appear that gender has a direct effect on rank, but it does appear that female experience has less of a positive impact on rank than male experience even after controlling for research output. It is possible that career interruptions may explain this differential effect, but the data do not allow us to identify such interruptions.

It appears that coauthored articles are not discounted much, if at all, relative to single-authored work.(22) To test the hypothesis that single-authored and coauthored articles are weighted equally as determinants of rank, we re-estimate equation (4) restricting the coefficients on the variables measuring number of single-authored and coauthored articles to be equal. Based on a Chi-square test of significance, we cannot reject the hypothesis that the two coefficients are equal at any conventional level of significance.(23) Moreover, we do reject at the 5 percent level the hypothesis that the weighting of coauthored articles is 1/n.(24)

In his recent article, Sauer reports results of a salary regression where the return to a coauthored paper with n authors is approximately 1/n times the return to a single-authored paper. Sauer's estimates are obtained from a data set consisting of academic economists who are Associate or Full Professors and who are employed by seven of the "top forty" departments. In contrast, our results on promotion decisions suggest that variations in academic rank do not reflect a 1/n weighting and, in fact, the results suggest that departments tend to count total articles rather than measure independent contributions of coauthors. The constrasting findings may result from differences in data and empirical focus as Sauer is interested in dollar returns to research to individuals who have achieved the highest academic rank at top schools. Our estimates are based on longitudinal data that contain observations over the first ten years of an academic's career and therefore include observations for all academic ranks across a broad range of academic institutions.(25)

Our data provide evidence of a potential bias in the promotion process in that rank determinations do not appear to distinguish between coauthored and single-authored publications. This failure to weigh each author's marginal contribution is of particular concern to female economists who may not be able to accumulate as many "lines" of total articles as men due to coauthorship barriers.(26)

IV. CONCLUSION

We have documented that economists tend to coauthor with others of the same gender. Moreover, female economists are less likely to coauthor in general than are males. While the effect of gender-sorting on female economists is somewhat mitigated by their tendency to self-select into larger departments, the implications of gender-sorting are not neutral with respect to promotion. In making promotion and salary determinations, a weighting factor of 1/n to measure contributions of coauthors would tend to maximize research output. However, we reject the hypothesis that 1/n weighting is used and, in fact, are unable to reject the hypothesis that single-authored and coauthored articles carry the same weight in determining promotion. In contrast, Sauer finds evidence of 1/n weighting in his study of salaries at top universities. This is not the case for promotion decisions in our broader sample of universities.

While the data place limitations on interpretation, gender-sorting in publishing, coupled with evidence of a "premium" being placed on coauthorship in promotion explains in part why women economists are not as likely to be promoted as men and wait longer for promotion. Our results raise a number of questions that await future investigation. Among these are questions regarding the impact of gender-sorting on research productivity. While the evidence suggests coauthorship neither enhances (nor reduces) research productivity, this result is based on a sample of individuals who have made publication decisions given existing promotion incentives. The implication is that female research productivity may be measured with a bias and that if faced with similar opportunities for coauthorship as males, female productivity may be enhanced.

TABLE I

Gender Distribution of Research Teams Consisting of Two or More Coauthors

 Gender of Coauthor
Gender of Sampled Total Number
 Individual Male Female of Coauthors
Male 119 8 127
 (%) (93.70) (6.3) (100.0)
Female 61 30 91
 (%) (67.03) (32.97) (100.0)

Total Number

 of Coauthors 180 38 218
 (%) (82.57) (17.43) (100.0)

Note: If the individual selects two coauthors (e.g., one male and one female) the selection generates two observations.

TABLE II

Logit Analysis of the Decision to Coauthor

(Single-authored = 0 Coauthored = 1)(a)

 (1) (2)
Intercept -2.822 -2.842
Experience 0.091** 0.170***
 (2.253) (3.158)
Department Size 0.115 0.088
 (0.681) (0.512)
Field Size 0.536** 0.263
 (2.018) (0.783)
Field Growth 12.559** 13.315**
 (2.365) (2.474)
Field Change 0.286 0.245
 (0.910) (0.781)
Gender -2.438* -2.807*
 (-1.947) (-1.929)
Gender*Department Size 0.601* 0.706**
 (1.929) (2.247)
Gender*Field Size 0.706
 (1.393)
Gender*Experience -0.186**
 (-2.278)
Log likelihood function -294.0(b) -290.7(b)

N = 464 Event = 0 273 Event = 1 191

Note: Student t-statistics in parentheses below coefficients

* = significant at .10 level, two-tailed test

** = significant at .05 level, two-tailed test

*** = significant at .01 level, two-tailed test

(a)An observation occurs each time an article is published. The individual and job-specific variables are measured as of the time of the publication as explained in Appendix A.

(b)[X.sup.2] statistic is significant at the .001 level.

TABLE III

The Estimated Impact of Coauthorship On Research Productivity(a)

 (3)
Intercept 1.154
Percent Coauthored -1.026
 (-1.205)
Publication Quality 0.428
 (0.194)
Age -0.105
 (-1.972)*
Gender -1.136
 (-2.029)**
Academic Experience 0.304
 (2.760)**
Department Size(b) 1.277
 (4.331)***

Note: Student t-statistics in parentheses.

* = significnt at .10 level, two-tailed test

** = significant at .05 level, two-tailed test

*** = significant at .01 level, two-tailed test

(a)The dependent variable, Weighted Articles, equals the weighted number of published articles at the end of ten years of experience. The coauthored articles are weighted by the number of authors.

(b)Department Size equals the mean department size measured over the years of academic experience.

TABLE IV

Bivariate Probit Estimate of the Relative Impact of

Coauthored and Single-Authored Research

on Academic Rank

(Assistant Professor = 0; Associate or Full Professor = 1)(a)

 (4)
Intercept -2.235
Number Single-Authored 0.236***
 (6.644)
Number Coauthored 0.191***
 (5.346)
Publication Quality 0.058*
 (1.686)
Institution Quality -0.178***
 (-5.618)
Experience 0.320***
 (4.610)
Gender 0.736
 (0.273)
Gender*Experience -0.141***
 (-3.073)
Log likelihood function -1418.3(b)

Note: Student t-statistics in parentheses below coefficients

* = significant at .10 level, two-tailed test

** = significant at .05 level, two-tailed test

*** = significant at .01 level, two-tailed test

(a)The dependent variable is Rank. Each individual in an academic setting is observed each year for each of ten years and the independent variables are measured eachyear as explained in Appendix A.

(b)[X.sup.2] statistic is significant at the .001 level. [Figure 1 Omitted]

APPENDIX A

Variable Definitions

Variable Definition (Source)
Academic Experience Years of experience at an academic institution (AEA Dire
ctory).
Age Age at year of receipt of Ph.D. (AEA Directory).
Coauthored Equals one if an individual's publication is coauthored,
 otherwise equals
 zero (AEA Index of Economic Articles).
Department Size Natural log of the number of AEA members affiliated with
 author's academic
 institution at the date of the observation. The measure
includes AEA members
 associated with departments other than economics and inc
ludes graduate
 student members. The variable is measured at years 1969,
 1974, 1978, 1981
 and 1985 with linear interpolated values between these y
ears (AEA
 Directory).
Experience Year of observation minus year of Ph.D. (AEA Directory).
Field Change Equals one if individual's primary field of specialty ch
anges over the ten
 year period following the year of Ph.D., otherwise equal
s zero (AEA
 Directory).
Field Growth Average annual growth rate (of AEA members) over the yea
rs 1969 to 1985 in
 the author's declared primary field of specialization AE
A Directory).
Field Size The number (000) of AEA members in author's declared pri
mary field of
 specialization as measured at years 1969, 1974, 1978, 19
81, and 1985 with
 linear interpolated values between these years (AEA Dire
ctory).
Gender Equals one if observed author is female, otherwise equal
s zero.
Institutional Quality The number of AER-equivalent pages per economics departm
ent faculty member
 at the observed author's academic affiliation. The varia
ble is measured at
 each year of experience. (Grave et al. [1982, Table 2]).
Number Coauthored Cumulative number of coauthored articles published at ti
me of observation.
 (AEA Index of Economic Articles).

Number Single-Authored Cumulative number of single-authored articles published at time of

 observation (AEA Index of Economic Articles).
Percent Coauthored For those who publish, the percentage of articles publis
hed at the end of
 ten years of experience that are coauthored (AEA Index o
f Economic
 Articles).
Publication Quality The mean number (00) of citations to the journal in whic
h the observed
 author's article(s) is published. The variable is measur
ed at each year of
 an individual's experience. (ISI, Journal Citations Repo
rts, 1985).
Rank Equals one if observed individual has achieved academic
rank of Associate or
 Full Professor, otherwise equals zero. The variable is m
easured at each year
 of an individual's experience. (AEA Directory).
Weighted Articles Cumulative number of articles published at the end of te
n years of
 experience, with coauthored articles discounted by the n
umber of authors
 AEA Index of Economic Articles).

APPENDIX B

Sample Statistics(a)

Variable Total Male Female
Academic Experience 6.29 6.64 5.93
 (4.20) (4.24) (4.15)
 [178] [89] [89]
Age 30.48 29.66 31.29
 (5.48) (3.31) (6.93)
 [178] [89] [89]
Department Size(b) 47.84 41.93 54.40
 (37.10) (36.51) (37.05)
 [95] [50] [45]
Field Change 0.10 0.12 0.08
 (0.30) (0.33) (0.27)
 [178] [89] [89]
Field Growth 0.05 0.06 0.05
 (0.02) (0.02) (0.02)
 [178] [89] [89]
Field Size 1.57 1.54 1.61
 (0.42) (0.44) (0.40)
 [178] [89] [89]
In Academic Institution 0.57 0.62 0.52
 (0.50) (0.49) (0.50)
 [178] [89] [89]
Institution Quality 13.38 13.29 13.47
 (13.28) (14.50) (11.95)
 [95] [50] [45]
Number Coauthored 1.51 1.83 1.19
 (2.28) (2.60) (1.86)
 [178] [89] [89]
Number Single-Authored 2.20 2.57 1.83
 (2.56) (2.79) (2.27)
 [178] [89] [89]
Percent Coauthored 35.84 36.26 35.38
 (31.30) (30.90) (31.99)
 [178] [89] [89]
Publication Quality 14.20 14.07 14.39
 (17.47) (17.14) (17.97)
 [661] [392] [269]
Rank 0.49 0.60 0.38
 (0.50) (0.49) (0.49)
 [178] [89] [89]
Total Number Articles 3.71 4.40 3.02
 (4.03) (4.50) (3.37)
 [178] [89] [89]
Weighted Articles 2.92 3.44 2.40
 (3.14) (3.47) (2.69)
 [178] [89] [89]

(a)The table reports means measured at the end of ten years of postdoctorate experience. Numbers in parentheses are standard deviations and numbers in brackets indicate sample size.

(b)The means reported are actual numbers of department AEA members.

(1)Findings on the gender-related wage gap are corroborated by a number of researchers using different samples and different time periods. As examples, see: Johnson and Stafford [1974]; Bayer Astin [1968]; Ferber and Kordick [1978]; Kenny and Studley [1989]; Barbezat [1987]. A recent study by Ransom [1989] suggests monopsonistic power of universities can explain differences in salaries. He estimates 10-15 percent of the gender gap is due to lower mobility of female versus male professors. (2)A number of studies examine earnings as a function of research output (articles) and other career attributes. See Diamond [1986], Hamermesh et al. [1982], Hansen et al. [1978], Siegfried and White [1973], and Tuckman and Leahey [1975] and most recently, Sauer [1988]. (3)As Weiss and Lillard note, lower productivity possibly is due to a reduced incentive to publish, perhaps arising from the lower likelihood of promotion. (4)As noted by a referee, possible explanations include preference and prejudicial attitudes, arising perhaps from "jealous spouses" and "manly men." (5)The top-twenty institutions are identified in Graves et al. [1982] and consist of Chicago, Harvard, Stanford, Wisconsin, University of Pennsylvania, MIT, Yale, UCLA, University of California at Berkeley, Princeton, Northwestern, Michigan, Washington, Rochester, Illinois, UNC-Chapel Hill, Columbia, NYU, Ohio State, and Minnesota. The cohort sample was selected by first collecting the name of each woman who received a Ph.D. in economics from one of the top-twenty institutions during the period 1968 to 1975. The Directory of Members of the American Economic Association (various volumes) was then referred to for demographic and career-related data. As needed, missing information was obtained by telephoning or writing to the individual. If data were available for both the beginning of the career as well as for a period of ten years of post-Ph.D. experience, the individual was included in our sample. A male cohort was then selected from the same Ph.D.-granting institution's graduating class. The male whose name alphabetically either preceded or followed (alternating every other time) the sampled female's was selected for sample inclusion so long as demographic and career related data were also available for the male. (6)To determine whether the selection is randomly distributed across gender, we evaluate the statistical difference between actual and predicted gender combinations based on the hypothesis that the sex of author and coauthor are independent. The Chi-square statistic of 25.64 indicates that the difference between actual and predicted gender combinations is significant at the .01 level. (7)While most studies that examine earnings as a function of article output fail to distinguish between single-authored and coauthored articles, Sauer makes a distinction. He finds, for his sample of economists at top schools, that the dollar return to a coauthored article with n authors is approximately [a/n.sup.th] the return to a single-authored article. (8)Women economists represent 16.5 percent of the total faculty in undergraduate institutions and 9.3 percent in graduate institutions. (Report of the Committee on the Status of Women in the Economics Profession, AEA Papers and Proceedings, May 1990). (9)Lindsay and Maloney [1988, 655]. (10)See McDowell and Melvin [1983], and Barnett, Ault and Kaserman [1988]. The trend toward coauthored research is not unique to economics. See, for example, Stephan and Levin [1987] who study six subfields in the areas of physics and earth sciences. (11)A fully-interacted model yields estimates similar to those appearing in the table. (12)The nonsignificance of the department size variable suggests department size is not a serious constraint for male researchers in the decision to coauthor. (13)See McDowell and Melvin [1983]. (14)It is of course possible that more recent cohorts of academic economists display different collaborative behavior than the cohorts in our sample. We cannot test this directly since complete ten-year publishing histories are not yet available for recent Ph.D. recipients. (15)These findings are also consistent with the idea that academic environments affect the "style and content" of academic research. Both Breit [1987] and Johnson and Johnson [1978], for example, argue that environment, inclusive of social relationships and personalities, plays an important part in the creation of new schools of thought ("Virginia School" and the "Keynesian revolution"). It is not surprising that, if such factors affect the content of economic research, then they can also affect the collaborative activity leading to publication. (16)Coauthorship may also affect quality of articles. Associated with this is the possibility that researchers select the ablest economists with whom to coauthor. However, our data do not indicate a statistically significant higher level of mean quality of coauthored versus single-authored articles. (17)Appendix B shows the means of the variable measuring weighted articles. These are 3.44 articles for men and 2.40 for women. (18)See Weiss and Lillard [1982]. (19)The percentage of coauthored articles for men is 36.3 percent compared to 35.4 for women (Appendix B). (20)The maximum likelihood estimation approach was introduced by Van de Ven and van Praag [1981]. (21)Nine individuals (three women and six men) are excluded from the sample because they were employed by foreign institutions for which data on institution quality and department size were not available. To explain whether an individual was in an academic setting we controlled for experience, gender, gender-experience, interaction, age, year of Ph.D. and its interaction with gender, total number of published articles, and whether an individual's primary field was in economic theory, economic history, or history of economic thought. The results indicate that being an academic is positively related to total number of articles published, having one's primary field in theory or history, gender, and with more recent Ph.D. cohorts. Being academic is negatively related to experience, year of Ph.D. interacted with gender, and age. The correlation between model disturbances is 0.620 (t=2.154). (22)Liebowitz and Palmer [1983] report results of a survey of department chairs which indicates that chairs generally assign a weight to coauthored papers that exceeds 1/n. Half of the respondents indicated they used weights in the 67%=75% range. (23)Using the Wald test, the statistic is [x.sup.2](1) = 0.984. (24)The Chi-square statistic is 3.869. (25)There are fifty-eight employer institutions represented in our sample (ranked from 1 to 231 by Graves et al. [1982]). We do find evidence that suggests higher ranking institutions evaluate single-authored articles more favorably relative to coauthored articles than do lower ranking institutions when we estimate the rank models with institutional quality interacted with number of single-authored and number of coauthored articles. This provides a further rationale for women selecting employment in larger institutions since there is a high correlation between large departments and high institutional quality. The correlation coefficient between the department size and institutional quality variable is 0.58. (26)Departments may elect not to discount work due to perceived positive externalities created by coauthoring. For example, coauthorship activity may lead to a more stimulating academic environment. If women tend to be excluded from collaborative teams, however, the policy implication is not necessarily one of reducing incentives to coauthor. One referee points out that perhaps women should be promoted with fewer publications than men, since lower output appears to be due to reduced coauthoring rather than reduced talent.

REFERENCES

Barbezat, Debra. "Salary Differentials by Gender in the Academic Labor Market." Journal of Human Resources, Summer 1987, 422-28. Barnett, Andy, Richard W. Ault and David L. Kaserman. "The Rising Incidence of Co-authorship in Economics: Further evidence." Review of Economics and Statistics, August 1988, 539-43. Bayer, Alan and Helen Astin. "Sex Differences in Academic Rank and Salary Among Science Doctorates in Teaching." Journal of Human Resources, Spring 1968, 191-200. Breit, William. "Creating the |Virginia School': Charlottesville as an Academic Environment in the 1960s." Economic Inquiry, October 1987, 645-57. Cole, Jonathan R. and Stephen Cole. Social Stratification In Science, Chicago: The University of Chicago Press, 1973. Diamond, Arthur M., Jr. "What Is a Citation Worth?" Journal of Human Resources, Spring 1986, 200-15. Farber, Stephen. "The Earnings and Promotion of Women Faculty: Comment." American Economic Review, March 1977, 199-206. Ferber, Marianne A. and Betty Kordick. "Gender Differentials in the Earnings of Ph.D.s." Industrial and Labor Relations Review, January 1978, 227-38. Ferber, Marianne A. and Michelle Teiman. "Are Women Economists at a Disadvantage in Publishing Journal Articles?" Eastern Economics Journal, August-October 1980, 189-93. Graves, Philip E., James R. Marchand and Randall Thompson. "Economics Departmental Rankings: Research Incentives, Constraints, and Efficiency." American Economic Review, December 1982, 1131-41. Hamermesh, Daniel S., George E. Johnson and Burton A. Weisbrod. "Scholarship, Citations and Salaries." Southern Economics Journal, October 1982, 472-81. Hansen, W. Lee, Burton A. Weisbrod and Robert P. Strauss. "Modeling the Earnings and Research Productivity of Academic Economists." Journal of Political Economy, August 1978, 729-41. Johnson, Elizabeth S. and Harry G. Johnson. The Shadow of Keynes. Chicago: The University of Chicago Press, 1978. Johnson, George E. and Frank P. Stafford. "Lifetime Earnings in a Professional Labor Market. Academic Economists." Journal of Political Economy, May/June 1974, 549-69. Kenny, Lawrence and Rogert Studley. "Economists' Salaries and Lifetime Productivity: What Have You Done For Me Lately?" Working paper, Stanford University, October 1989. Liebowitz, Stanley J. and John P. Palmer. "Assessing Assessments of the Quality of Economics Departments." Working paper, University of Rochester, December 1983. Lindsay, Cotton M. and Michael T. Maloney. "A Model and Some Evidence Concerning the Influence of Discrimination on Wages." Economic Inquiry, October 1988, 645-60. McDowell, John M. "Obsolescence of Knowledge and Career Publication Profile: Some Evidence of Differences Among Fields in Costs of Interrupted Careers." American Economic Review, September 1982, 752-68. McDowell, John M. and Michael Melvin. "The Determinants of Coauthorship: Analysis of the Economics Literature." Review of Economics and Statistics, February 1983, 155-60. Ransom, Michael. "Monopsony as a Source of Gender Differentials in Pay in the Academic Labor Market." Working paper, Brigham Young University, March 1989. Sauer, Raymond. "Estimates of the Returns to Quality and Co-authorship in Economic Academia." Journal of Political Economy, August 1988, 855-66. Siegfried, John J. and Kenneth J. White. "Financial Rewards to Research and Teaching: A Case Study of Academic Economists." American Economic Review, May 1973, 309-15. Stephan, Paula E. and Sharon G. Levin. Demographic and Economic Determinants of Scientific Productivity, November 1987. Tuckman, Howard P. and Jack Leahey. "What is an Article Worth?" Journal of Political Economy, October 1975, 951-67. Van de Ven, W. P. M. M. and B. M. S. van Praag. "The Demand for Deductibles in Private Health Insurance." Journal of Econometrics, November 1981, 229-52. Weiss, Yoram and Lee Lillard. "Output Variability, Academic Labor Contracts, and Waiting Times for Promotion," in Research in Labor Economics, edited by Ronald G. Ehrenberg. Greenwich: JAI Press, 1982, 157-88.

JOHN M. McDOWELL and JANET KIHOLM SMITH, Department of Economics, Arizona State University. We wish to thank Ray Sauer, Stuart Low, Art Blakemore and three referees for helpful comments on earlier drafts.