Faculty Research Productivity: Why Do Some of Our Colleagues Publish More than Others?

The justification for studying faculty research productivity is that it affects individual advancement and reputation within academe, as well as departmental and institutional prestige (Creamer 1998, iii). Publication records are an important factor in faculty performance evaluations, research grant awards, and promotion and salary decisions. The phrase “publish or perish” encapsulates the importance of research productivity to academic careers. In addition, questions are sometimes raised about whether an individual's status as a minority within academia (e.g., being a member of an underrepresented ethnic or racial group or being female in a male-dominated profession) affects his or her ability to publish or likelihood of publishing (Cole and Zuckerman 1984; Bellas and Toutkoushian 1999). Finally, most previous work that tackles the productivity causality puzzle comes from disciplines other than political science.¹ Thus, one of the purposes of this report is to explore whether the existing findings about research productivity in other disciplines apply equally well to research productivity in political science.²

The question that we wish to answer is: What factors contribute to higher or lower research output by political scientists? We base our answer on responses to a 2009 survey sponsored by the APSA. Respondents were drawn from a sample of all faculty employed in political science departments (including departments of government and public affairs) throughout the United States. (Appendix A provides a description of the survey methodology.)

According to previous studies, several blocks of variables determine scholarly productivity. These variables include demographics and family-related factors, human capital, opportunity costs (teaching and service workload), working environment, and professional variables (table 1). Among the demographic variables listed in table 1, gender differences have received special attention. Numerous studies have revealed that women publish less than men (Fish and Gibbons 1989; McDowell and Smith 1992; Broder 1993, 123; Bellas and Toutkoushian 1999; Sax et al. 2002; Maske, Durden, and Gaynor 2003, 561; Taylor, Fender, and Burke 2006; Evans and Bucy 2010). This finding, however, remains controversial: Davis and Patterson (2001, 89) argue that women do not publish significantly less than men when source of Ph.D., type of employer, and field of specialization are held constant.

Table 1 References for Explanatory Variables for Scholarly Productivity

The second category of variables found in table 1 concern “human capital.”³ Human capital addresses any contextual or individual attributes that could potentially influence the quality of an individual's research skills or training. The professional reputation of an academic's Ph.D.-granting department is consistently tied to differences in research productivity (Hansen, Weisbrod, and Strauss 1978; Davis and Patterson 2001, 88; Broder 1993; Buchmueller, Dominitz, and Hansen 1999, 71). The assumption is that top-rated schools attract the best students and then provide them with training at the frontiers of the discipline and socialization into a culture that values high-quality research (Rodgers and Neri 2007, 76).

“Opportunity cost” variables capture the time spent teaching or doing service. Given the limited amounts of time that faculty have, teaching or administrative requirements set by the employing institution may affect faculty research productivity (Fender, Taylor, and Burke 2005; Taylor, Fender, and Burke 2006; Maske, Durden, and Gaynor 2003). Studies consistently reveal that a large teaching load significantly reduces published output (Graves, Marchand, and Thompson 1982).⁴

The category of “current working environment”—both its culture and its availability of resources—captures primarily departmental and institutional characteristics. Broader availability of resources and incentives for publishing should influence publication rates (see table 1).⁵ “Culture” relates to shared attitudes about not only the value of research, but also collegiality and interpersonal encouragement. Each academic's own research productivity is affected by the productivity of his or her departmental colleagues through “collaboration, academic discourse, peer expectations [and] peer pressure” or through colleagues' other attributes, such as “ability, integrity [and] professionalism” (Rodgers and Neri 2007, 85; see also Taylor, Fender, and Burke 2006).

As a category that is distinct from the working environment, we also consider “professional variables,” which include the achievements of an individual's academic career. For example, scholarly productivity has been associated with the ranking of the program with which an individual is affiliated (Davis and Patterson 2001, 88; Xie and Shauman 1998, 865; Garand and Graddy 1999; McCormick and Rice 2001; Youn 1988). It may be that higher ranked departments select better scientists, or perhaps these departments foster greater productivity (Broder 1993, 116).⁶ Arguably, faculty research productivity also varies according to the researcher's specific subject matter (Fish and Gibbons 1989, 98).

Faculty rank (instructor/lecturer, assistant professor, associate professor, full professor) is considered a professional variable. Some researchers find rank to be a predictor of productivity (Blackburn, Behymer, and Hall 1978; Bellas and Toutkoushian 1999; Dundar and Lewis 1998; Sax et al. 2002; Xie and Shauman 1998, 865), while others have shown that rank has no influence on faculty research productivity when other relevant variables are taken into consideration (Over 1982; Wanner, Lewis, and Gregorio 1981).⁷ Also categorized under professional variables is coauthorship, which is thought to “increase article production through the division of labor made necessary by increased complexity in the subject matter and by the need to saturate markets to increase the probability of getting papers accepted for publication” (Maske, Durden, and Gaynor 2003, 555, 561; see also Hollis 2001; Durden and Perri 1995; Davis and Patterson 2001, 90; Taylor, Fender, and Burke 2006).

The theories behind the explanations for variation in research productivity are nearly as varied as the factors studied. Behavioral reinforcement theory views the “system of faculty ranks as a reward system as well as a schedule of reinforcement” (Tien and Blackburn 1996, 5). A similar idea is proposed by the investment-motivated model of scientific productivity, which argues that “scientists engage in research because of the future financial rewards associated with the activity” (Levin and Stephan 1991, 115). Such a model implies a decline in research productivity over the course of an individual's career, given the finite time horizon (Diamond 1984). Rodgers and Neri (2007, 79) report that the most productive period is the first five years after the Ph.D. is conferred, and Davis and Patterson (2001) report that productivity generally declines after tenure.

In contrast, a consumption-motivated model that stresses the “scientist's fascination with the research puzzle itself” (Levin and Stephan 1991, 115) does not predict a decline in research productivity over time. Likewise, selection theory (Finkelstein 1984) argues that only the most productive faculty members are promoted, eliminating low producers before they reach higher ranks and thus creating a situation in which higher ranking faculty produce more. Accumulative advantage theory emphasizes the importance of resource acquisition over time (Allison and Stewart 1974). Motivational theory draws an important distinction between intrinsic motivation (e.g., interest in research) and extrinsic motivation (e.g., desire for promotion). Intrinsic motivation may account for the continued productivity of full professors, who are no longer motivated by the possibility of promotion (Finkelstein 1984, 101).

Thus, theoretical approaches to the productivity question vary, as do the factors that predict faculty research productivity. The dependent variable (productivity, or research output) can also be evaluated in a variety of ways. In our analysis of scholarly productivity, we use as our dependent variable the respondent's best estimate of the total number of articles that he or she has published in refereed academic or professional journals over his or her entire career.⁸ We also look at a summary measure that includes refereed journal articles, books, edited books, and book chapters. Finally, we create a model that uses books and book chapters published as controls when evaluating the total number of articles published. (See appendix A for a description of all variables included in the analyses.)

Descriptive Statistics

Before we embark on the multivariate analyses, we first provide descriptive statistics. Across the entire set of 1,399 respondents, the average number of articles published in refereed academic or professional journals during the respondent's entire career is 10.5. In table 2, we divide our sample into groups of men and women and compare their publication rates. On average, men publish significantly more articles than women do. We wish to note that this bivariate calculation does not control for age. The average age of women in the profession is lower than the average age for men in the profession. Thus, on average, men in the profession have more years of publishing time (based on age) than have women.⁹

Table 2 Average Number of Articles Published by Subgroup

Notes. For all, p < .000.

^a Percentages exclude one transgendered respondent.

^b Percentages do not add to 100% because of the exclusion of respondents from Tier IV, unranked, and unknown departments.

^c Percentages do not add to 100% because of the exclusion of respondents from departments within a two-year college and respondents without a program type specified.

^d Percentages do not add to 100% because of the exclusion of instructors, lecturers, postdocs, fellows, and respondents without a rank specified.

Turning to human capital variables, table 2 shows that the average number of articles published during an individual's career is significantly higher among graduates from departments ranked among the top 25 (tier I schools) than among graduates from departments ranked 26–50 (tier II) and departments classified as tier III using the Schmidt and Chingos ranking (2007).¹⁰ Thus, the ranking of an academic's Ph.D.-granting department is bivariately tied to differences in research productivity.

Looking at the opportunity cost variables, we asked respondents to report their typical teaching load each year (for the past five years). Across all respondents, the average number of courses taught is 4.3 per year. For purposes of a simple descriptive picture, we divided respondents into three groups: faculty with low teaching loads (2.5 courses or less per year), medium teaching loads (3 to 5.5 courses per year), and high teaching loads (6 or more courses per year). On average, faculty members with the lowest teaching loads publish 14.5 articles, while individuals with heavy teaching loads publish 4.9 articles (see table 2). These numbers reveal a major difference in research output depending on how many courses a faculty member teaches.

The last two variables that we considered in table 2 are associated with working environment and professional achievement. Looking at the “current employment” rows, we can see that faculty who are employed by Ph.D.-granting departments publish dramatically more than faculty who are employed by MA-granting programs or departments in a four-year college. The numbers in table 2 also illustrate the effect of the professional achievement variables by highlighting the difference in the number of articles published by members of different professional ranks. On average, assistant professors have published 3.6 articles, while full professors have published 18.6 articles.

Analysis

In running OLS regressions in our multivariate analysis, we employed three different versions of the dependent variable. The first simply used the respondent's raw report of the total number of articles published in refereed academic or professional journals over his or her entire career. The second approach replaced missing responses to this question with the value of zero. The third approach followed a recommendation by Fox and Milbourne (1999, 256) that the number of articles published be transformed as the logarithm of one plus the original variable, with missing responses replaced with zero. This helps deal with a concentration of observations at zero and makes the distribution more closely approximate a normal distribution.¹¹ The tables that we publish here all use the logarithmic transformation of the number of articles produced as the dependent variable. Analyses using the other two versions of the dependent variable are consistent with the findings reported here and are available from the authors upon request.

Table 3 reports regression results using the logarithm of one plus the number of articles published (with missing responses replaced with zero) as the dependent variable. Model I of table 3 contains only demographic and family-related factors.¹² These factors explain 5% of the variation in the (log of the) number of articles produced. Two variables emerge as significant predictors: gender and number of children. According to this simple demographic model, women tend to publish less than men. Additionally, as the number of children that an individual has increases, so does the number of articles that he or she publishes.

Table 3 Log of Articles as Dependent Variable, Based on Original Responses, with Missing Responses to Predictor Variables Excluded

Note.

*** p < .001,

** p < .01,

* p < .05

Model II of table 3 incorporates the measures of human capital. Both Ph.D. program rank and number of years to complete the Ph.D. are significant. As the ranking of the program from which a faculty member received his or her Ph.D. improves, the number of articles this individual publishes increases.¹³ As the number of years to complete the doctoral degree increases, the number of articles published decreases.

The problem that appears in table 3, however, concerns Model I's loss of a large number of respondents because they neglected to report their number of children. The introduction of additional variables in Model II results in a further loss of 225 other cases. This loss occurs in part because 141 respondents did not identify the institution from which they received their Ph.D.; this prevents us from using the 2007 Schmidt and Chingos ranking variable for these respondents. The more significant missing value problem arises with the question: “In what year did you obtain your Ph.D. degree?” Within our sample, 273 respondents either did not answer or made a mistake when typing in a year. Given this missing data problem, we decided to use the multiple-imputation Amelia II program for missing data (Honaker, King, and Blackwell 2010).¹⁴ We used this program to impute estimates of the missing responses on each of the independent variables used in the analysis. The purpose of this approach was to increase the number of observations taken into consideration in the analysis.

Table 3.1 reports the testing of exactly the same models as are tested in table 3, but using the imputed data and holding the number of cases in the analysis at 1,399 for all models. As in table 3, both human capital variables are significant in Model II of table 3.1. When using the imputed data, it is noteworthy that the significant demographics include “married.” Thus, table 3.1 reveals that women tend to publish fewer articles than men, while married and partnered persons publish more than single, divorced, or widowed academics.

Table 3.1 Log of Articles as Dependent Variable (Regressions with Imputed Data)

Note.

*** p < .001,

** p < .01,

* p < .05

In Model III of table 3, we add in the controls that we label opportunity costs: teaching load, number of new courses prepared, number of committee memberships, number of committees chaired, and amount of student advising. In this model, we jump to 29% of the variance explained in table 3 (Model III) but lose another 108 cases. According to these results, heavy teaching loads do take their toll on article production. In addition, the larger the number of new courses prepared, the lower the number of articles published. We also find, rather unexpectedly, that the more committees an individual chairs and the more advisees he or she has, the more he or she publishes.

When all respondents are considered and missing values are replaced by imputed data, the strong relationship between a higher teaching load and lower article production is confirmed, as is the negative relationship between new course preparation and article production (Model III in table 3.1). The overall count of advisees is again positively related to the number of articles published. To explain the findings regarding committees and advising, it may be wise to interpret both as measures of professional involvement rather than opportunity costs. It appears that we misclassified advising and committee service as opportunity costs, given that when these variables are significant, they appear to support rather than detract from article production. We return to this finding in our discussion section.

Model IV includes the working environment variables: collegial climate; count of overall resources; and current employment in a Ph.D. program, MA program, or private institution. Model IV in table 3 reveals that more positive evaluations of a department's “climate” are related to fewer articles published. Faculty members who evaluate their environment as more friendly, respectful, collegial, collaborative, and cooperative publish less, on average, than faculty who evaluate their home department as more hostile, disrespectful, contentious, individualistic, and competitive. This relationship emerges even more powerfully in the imputed data (Model IV, table 3.1).¹⁵ The negative association between departmental collegiality and research productivity is affirmed in the tables that follow. We must accept the finding that faculty members who operate in the more competitive, individualistic, and hostile departments publish more on average. The defining element of this scale is the collegial versus contentious contrast, with collegiality associated with lower total publications and contention associated with a higher number of publications.

We also see from the test of Model IV (in table 3 and table 3.1) that more resources are associated with an increase in the number of articles published. The dummy variable for employment in a Ph.D.-granting department is also significant. Table 3 and table 3.1 show contradictory results regarding whether faculty members in private institutions tend to publish more than faculty members in public institutions. Table 3.1's finding that faculty members employed in private rather than public institutions tend to publish more is consistent throughout the analyses using imputed values for missing responses (and therefore based on the largest possible number of respondents). Note that with Model IV, we can now explain 41% of the variance in the number of articles published (see table 3.1).¹⁶

The last model that we report in tables 3 and 3.1 includes the following professional variables: current faculty rank, a series of dummy variables for current primary field of teaching and research, year the Ph.D. was granted, coauthorship, frequency of conference attendance, and the ranking of the department that currently employs the faculty member. This final model is extremely powerful; it explains 44% of the variance in the number of articles published.¹⁷ Model V in tables 3 and 3.1 shows that as an individual moves up the academic ranks, the total number of articles published in his or her career also increases.¹⁸ We do not find evidence that faculty members in any subfield publish significantly more than faculty members in another subfield.¹⁹ We do find that as more time passes since the granting of the degree, the number of articles published increases.²⁰ Increased conference attendance is also positively related to greater article output (see tables 3 and 3.1).

We also tested Models III, IV, and V after adding the total number of books written or edited and book chapters published as a control. Our findings confirm the results of Maske, Durden, and Gaynor (2003, 561), who report a significant positive relationship between books published and articles published. Among our respondents, the bivariate correlation between the number of articles published and the total number of books, edited books, and book chapters is .640. The correlation between the number of articles published and the number of books published is .593. A higher number of books, edited books, and book chapters is positively associated with a higher number of articles published. Thus, the activities are clearly complementary and do not detract from each other; rather, they reinforce each other.

In a related vein, we thought it valuable to report on a different approach to measuring research output. As an alternate dependent variable, we evaluate the total number of publications. To create a total publications variable, we add together responses to four questions: (1) number of articles published in refereed academic or professional journals, (2) number of monographs (books) published, (3) number of books edited, and (4) number of book chapters published. For the results presented in table 4, missing responses to all questions are set to zero and the responses are summed (the value of one is added to the sum before the log is calculated). To save space, we do not present the results for Models I, III, and IV, and for Model II, we report only the results using the data file created by Amelia, with missing responses on the independent variables replaced with imputed values.

Table 4 Log of Total Productivity as Dependent Variable

Note.

*** p < .001,

** p < .01,

* p < .05

For Model V of Table 4, we report results using the imputed data files created by Amelia, as well as results using the log of total productivity (plus one) as the dependent variable and allowing missing cases to be deleted from the analysis. If we focus on Model V of Table 4, which controls for human capital, opportunity costs, and important characteristics such as faculty rank, we see that women on average report lower total publications than men. Being a minority and having children are significant when using the imputed data, but when missing values are dropped from the analysis, these characteristics are not significantly related to overall total output.

According to table 4, the longer the time an individual spends earning the Ph.D., the lower his or her number of total publications. When opportunity costs are considered, we see that the more courses that are taught, the lower a faculty member's total number of publications. The results using imputed values to replace missing responses also reveal a strong relationship between more advisees and more publications. Among the working environment variables, the total number of publications tends to be lower when the department's collegiality is high. The total number of publications is higher among those faculty members who report receiving more of the following resources: course release time, research assistance time, discretionary funds, travel funds, and summer salary. Consistently significant predictors of total productivity among the professional variables are higher faculty rank, year of degree (i.e., more time since finishing the doctoral degree), and conference attendance.

Since promotion to a higher rank and the total number of publications are inextricably combined in academia, we thought it important to divide our sample into subgroups based on academic rank and evaluate the factors that predict different levels of productivity within each rank. Acknowledging that full, associate, and assistant professor are each heterogeneous categories, in table 5, we test Model V within ranks to identify why some faculty members are more productive than others.²¹ In table 5, we report the results of testing Model V among assistant professors only, associate professors only, and full professors only. Table 5 uses the log of one plus the number of articles published as the dependent variable. We report results based on the file created by the Amelia program, with missing responses replaced with imputed values, as well as results based on analyses with missing responses excluded from the calculations. The following findings focus only on those columns based on the data with missing responses replaced by imputed values, as we feel more confident in these results, given that they incorporate a larger number of respondents.

Table 5 Log of Articles as Dependent Variable in Subgroup Regressions

Note.

*** p < .001,

** p < .01,

* p < .05

Looking first at assistant professors only (table 5, Model VA), we find that a higher number articles published (with the logarithmic transformation and using imputed values for missing responses) is associated with being male rather than female, having more children, taking less time to complete the doctoral degree, teaching fewer undergraduate courses, having more resources, working in a private institution, and attending more conferences. Among associate professors only (table 5, Model VB), a higher number of articles published is associated with being male rather than female, graduating from a higher ranked department, teaching fewer undergraduate courses, working in a competitive rather than a collegial climate, having more resources, and being employed in a department with a Ph.D. program. Among full professors only (table 5, model VC), a higher number of articles published is associated with being male rather than female, having more children, having more advisees, working in a competitive versus a collegial department, having more resources, being employed in a Ph.D.-granting department, working at a private institution, having a specialty other than American politics, and more years since receipt of the Ph.D.

Thus, we can see both similarities and differences in the predictors of article publication rates across academic ranks. For example, the availability of more resources is related to more publication at all ranks. A large teaching load appears to have detrimental effects on publication rates for assistant and associate professors, although not for full professors. A larger advising load is associated with more productivity among full professors, but not among associate and assistant professors. As well, when we look within these subgroups (partially controlling for age), we find that working in a private institution is associated with higher publication rates among assistant and full professors but has no significant effect on the number of articles produced by associate professors. Being employed by a Ph.D.-granting department is positively related to article production among associate and full professors, but not among assistant professors.

Discussion

With regard to demographics, our results appear to reveal that women employed in political science departments in the United States publish less on average than their male counterparts. When we divide respondents by rank and conduct our analysis within these ranks, using imputed data, the relationship between articles published and gender is significant at all ranks. Several explanations for the existence of this gender difference have been offered: Xie and Shauman (1998), for example, argue that female scientists are less likely to hold the positions and have access to the facilitating resources that are conducive to higher rates of publication performance. This finding may be relevant for political scientists at the associate professor level, at which women are less likely than men to be employed by a top-ranked department. At the assistant professor level, however, women are more likely than men to be employed by a top-ranked department. In addition, women and men on average report equal access overall to resources.²²

Another explanation that has been offered in the literature is that women spend more time “mentoring” than do male faculty. Collins (1998) finds that women are more likely than men to devote time to teaching and advising, serve in part-time positions, and teach in fields unlike the ones in which they were trained. Among our respondents, we do not see significant differences between men's and women's teaching loads, either for graduate or undergraduate courses, nor do we see significant differences in committee membership, committee chairing, or overall levels of advising. We note, however, that our questions count the number of these activities but do not ask respondents to report on the amount of time spent on these activities. Female political scientists are also no more likely than male political scientists to work in part-time positions and no more likely to teach or do research in a field that differs from their major field as a graduate student.

Another explanation is that women spend more time than their male colleagues on household and childcare responsibilities (Gmelch, Wilke, and Lovrich 1986; Suitor, Mecom, and Feld 2001).²³ This explanation seems valid if traditional divisions of labor between men and women exist within the household. We did not, however, include a question in our survey about time devoted to domestic or child-rearing chores, so we cannot check this hypothesis. Looking at our sample of political science faculty members, we see that at the level of assistant professor, men are more likely than women to have children. At the associate and full professor levels, men and women are equally likely to have children.²⁴

It is also important to note that the men in our sample have, on average, been in the profession longer than the women in the sample. On average, female respondents received their doctoral degree in 1994, while male respondents received their degrees in 1990. However, we do control for year that a degree was awarded in Model V, and we still find in much of our analysis that women publish less than men.

The findings regarding whether members of a racial or ethnic minority publish more or less when opportunity costs, working environment, and professional characteristics are taken into consideration are inconsistent. We do not find any relationship between self-identification as a minority and number of article publications when the sample is divided by rank.²⁵ We note that among the political science faculty who responded to our survey, racial minorities are no more or less likely to be employed by a department that offers a doctoral or an MA degree. Political science faculty who are members of a racial or ethnic minority group are also no more or less likely to be married or have children, and they have, on average, the same number of children as do nonminorities. We find no differences on average in age or year that a degree was awarded when we compare minority group respondents with nonminorities. The average teaching load at the graduate level is the same for minorities and nonminorities, while at the undergraduate level, minorities have a slightly lower teaching load.

A finding that we think particularly important to a profession that places a great deal of emphasis on publications when evaluating faculty performance is the negative effect of a heavy teaching load on research output. The opportunity costs of teaching a large number of courses and preparing new courses are significant indeed. Thus, our findings correspond to the findings of many other scholars—that time spent teaching takes away from time spent doing research (Maske, Durden, and Gaynor 2003, 561; Bellas and Toutkoushian 1999; Xie and Shauman 1998, 865; Hamovitch and Morgenstern 1977, 636; Porter and Umbach 2001; Taylor, Fender, and Burke 2006, 858).

Our findings do diverge from previous findings regarding what is generally classified as “service.” A heavy teaching burden generally has a negative effect on publishing, but advising does not. We speculate that the positive relationship between student advising and higher article production is related to the constructive effects on intellectual activity (including the possibility of coauthorship) that are associated with frequent one-on-one interaction with advisees. We believe that advising represents a measure of professional involvement and should be considered a bonus rather than a cost. A particularly strong relationship exists between the number of Ph.D. students an individual advises and the total number of articles that he or she publishes in refereed academic or professional journals. Having a publishing research group and advising appear to go hand in hand.

Finally, we would like to highlight one more finding. The presence of a collegial climate within the department tends to be associated with less productivity. In other words, a degree of competiveness, even hostility, does not detract from productivity. Other attitudinal measures, such as one's evaluation of the research climate within the department, are positively associated with publications. We will address this evaluative dimension of the professional environment further in a follow-up report.