Reputation and earnings: the roles of quality and quantity in academe.
(Business Week, July 9, 2007)
There is a substantial literature in economics on the formation of reputation, both of individuals and of the groups to which they belong. Most of the research has been purely theoretical, with an analysis of reputation of firms (goodwill) as an asset (Tadelis 2002) and with much of the work focused on reputation arising from behavior in repeated games (Mailath and Samuelson 2006). Presumably, reputation, defined as "overall quality or character as seen or judged by people in general," is something that develops over time in the minds of those who are judging the person, group, product, and so on. (1)
In this study, we provide some evidence on the determinants of reputation using the example of academic economists. The specific questions are (1) How does the quantity of publications affect the regard in which a scholar is held by other scholars? (2) Do a few extremely well-regarded publications have the same reputational effect as an equally successful (in terms of its total impact on other scholars) but lengthier publication list? and (3) Are the determinants of reputation the same as the determinants of pecuniary returns? The answers to these specific questions about the rewards to scholars might shed some light on the general determinants of professional reputation. These questions do not appear to have been addressed in economics. The sparse literature on academic reputation in other disciplines has concentrated on solicitations of senior scholars' opinions about curricula vitae rather than on proxies for reputation that are based on actions (Feist 1997 on natural scientists and Hayes 1983 on psychologists).
II. THINKING ABOUT REPUTATION
Presumably, reputation is related to observers' memory and how the actions and sequences of others' behavior produce those memories. As such, the literature on memory and learning in experimental psychology may be informative here. That literature unsurprisingly makes it clear that memory is enhanced by additional exposure. In terms of our question about the trade-offs among the dimensions of quality, however, the issue is whether memory of an item within a class is better enhanced by a given number of stimuli of one item in that class or of several different items in the same class.
A fundamental work in the general area of memory (Tulving and Thomson 1973) demonstrated its complexities and proposed a theory of "encoding specificity"--that the specifics underlying exposure to events and the keys that might lead to the retrieval of memory interact to determine how memories develop. This study led to a huge literature, none of which speaks directly to our question, but part of which may shed some light on it. Arnold and Lindsay (2002) imply that people will remember better if they are stimulated by exact repetition of an event rather than by variations in it. Starns and Hicks (2005) show that providing related stimuli at the same time has complementary effects on memories of each, but that this is only true if the stimuli are provided in the same experimental session. In a slightly different context, that of studying for tests, the results of Kurtz and Loewenstein (2007) show qualitatively similar results. Overall, one might infer that these experiments support the notion that memory will be more strongly enhanced by repetitions of the same stimulus than by the same number of different, but related stimuli. Viewing the class of stimuli as all references to a scholar's work, and each individual stimulus as a reference to a particular study, they suggest that scholarly reputation may be more strongly affected by a very important publication than by an equally visible series of lesser works.
We examine in particular how the market for reputation, R, is affected by three kinds of stimuli produced by agents' efforts: quantity and two types of quality. (2) We define quantity as the number of publications, Q, and quality as [q.sub.1], the total recognition of a scholar's entire oeuvre by other scholars, and [q.sub.2n], the recognition of his/her nth-most recognized publication (where n is some arbitrary small number, with n = 1 in most of our empirical work). We assume that the members of a scholar's profession consider his/her outputs in creating their estimates of the scholar's reputation. Our scholarly works interact in the minds of those who determine our reputations according to some function:
(1) R = R(Q, [q.sub.1], [q.sub.2n]),
with [R.sub.j] [greater than or equal to] 0 for all j dimensions of reputation- enhancing activity.
The identity of those who determine one's reputation varies with the example. Certainly the phrase from the definition, "people in general," should not be interpreted too broadly. The public in general does not determine and probably does not care about the reputations of individual scholars. Also, within the scholarly community some members have a greater influence in contributing to judgments about reputation than others. Because of the difficulties in identifying members of the groups that determine reputation, and even, as the epigraph to this study suggests, of defining it, we base our example on a wide variety of proxies for a scholar's reputation.
III. TESTING THE DETERMINANTS OF REPUTATION--MEASURING INPUTS AND OUTPUT
One approach to measuring reputation is based on the notion that the awards that our peers confer on us reflect their assessments of our reputations. A second, indirect approach assumes that the members of an academic collective (department) seek to avoid diminishing its reputation. To achieve this goal they add only those marginal members whose individual reputations are at least as high as some summary measure of the reputations of the collective's current members (e.g., Basu 1989; Rosen 1987, and, by analogy, the literature on the worker-managed firm going back to Ward 1958). (3) Thus for members of collectives that are sufficiently large, a measure of individual reputation is the reputation of the collective. Taking the same tack, the collective's reputation is an even better proxy for the reputations of its newer members--those whose reputations were recently deemed by more senior members of the collective to be sufficiently high that their admission to the collective would not reduce its reputation.
Our sample consists of full professors in those economics departments that are included among the 88 U.S. departments listed as being in the top 200 in the world by Kalaitzidakis, Mamuneas, and Stengos (2003). This provides a sample of 1,339 scholars. Although this sample is obviously selected--its members had to have sufficient individual reputations to be included in this fairly elite group of institutions--there is presumably enough variation in reputation across the 88 departments that selectivity bias is not solely responsible for any results we obtain. (4)
To proxy individual reputation by awards conferred, we first define Honored as equaling one for those sample members who received a Nobel Prize, who were elected President of the American Economic Association (AEA), who were named a Distinguished Fellow of the AEA, or who received its Clark Medal (until 2009 awarded biennially to an economist under age 40). The set of individuals whose decisions produce these proxies for reputation is the Nobel Committee (with solicited advice) for the Nobel Prize and various AEA committees for the other awards.
The difficulty with the variable Honored is that economists are extremely stingy in providing distinctions to each other--using this definition, an award is received by only a tiny fraction of individuals even in this elite sample. A more broadly conferred indicator of reputation is election as a Fellow of the Econometric Society (a description of these elections is in Hamermesh and Schmidt 2003). Sample members account for all fellows who are not emeriti in the U.S. economics departments and for nearly half the fellows worldwide. This indicator for reputation is determined by votes of existing fellows.
The second series of measures is based on the reputation of the department with which the scholar is associated and stems from our hypothesis that the collective's reputation forms a lower bound of the individual's. The first measure is the department's ranking in Kalaitzidakis et al. (2003) (between 1, the highest, and 200, the lowest), based on publications by members of the department. We also use the U.S. News and World Report subjective ratings of graduate programs in economics as another proxy for reputation. The National Research Council's (1995) subjective ratings of faculty quality in 1993 are yet a third measure of departmental reputation. All three allow us to examine the determinants of the collective reputations of the institutions which chose to employ the scholars, thus implicitly proxying the lower bounds of each scholar's reputation. (5) That is clearly the case using the NRC93 ratings as measures of the reputations of people who moved into or were tenured in those departments after 1992.
None of the many proxies for reputation that we examine is perfect. One can surely find fault with any of them. But they are all reasonable proxies and are at least to some extent independent reflections of reputation. Thus, if we find consistent impacts of the measures of quality and quantity on these proxies, we can be fairly secure in drawing conclusions about their effects on reputation.
The measures of quantity and the dimensions of quality all come from the Web of Science [R] Social Science Citation Index, which includes articles in an immense variety of refereed and unrefereed journals, but no books or unpublished papers. (6) Social science journals are where economists publish most of their scholarly work and are thus the outlets in which they establish their reputations and in which other scholars acknowledge their influence. (7) To represent Q we use the number of entries in a scholar's record from 1956 to 2007, a period that encompasses the work lives of all members of our sample. The dimensions of quality are represented by citations, with [q.sub.1] proxied by the total citations to a scholar's works that are included in Q and [q.sub.2n] being the numbers of citations to the scholar's nth most-cited work, n = 1, ..., 5. (8) Obtaining information for those whose names are unique (like the authors of this paper) is not difficult; for nearly 100 others it required direct comparisons of curricula vitae to the entries in the SSCI and, in several cases, correspondence with individuals in the sample.
Are citations an input into reputation or are they reflections of reputation? Put differently, is a scholar's work cited because of the work's inherent quality or appeal, or because she/he is at a particular institution? There is no way of proving that the former answer is correct; but there is substantial evidence that our own behavior and the quality of our own work are determined by ourselves and that the profession does not pay attention to affiliation in judging scholarship (Blank 1991; Hamermesh 1994). It seems reasonable to treat citations as inputs into reputation and, as has been done in huge amounts of research, into salary determination.
The other measures account for individual characteristics that may affect reputation and that may be correlated with Q or one of the q. We include a quadratic in the year of the author's first publication listed in the SSCI to measure the time available to construct a reputation. The gender of the scholar may bias our measures of quality, as some have argued (Ferber 1986) that same-sex citation is a common practice, although others (Hamermesh and Schmidt 2003) fail to find any disparate outcomes in receipt of a particular award. The author's place in the alphabetical list in the sample may function similarly, as some (Einav and Yariv 2006) have shown that those whose names are earlier in the alphabet tend to be favored in certain aspects of scholarly work. We used personal knowledge or vitae to generate a measure of whether the person received his/her undergraduate education at an English-language university, on the grounds that migration to the United States from the upper tail of the ability distribution might occur (Borjas 1987) and might be correlated with the quantity and quality measures. Finally, we created an indicator for anyone who had published an article in the sub-field of theoretical econometrics. Most of the results use these controls, but including them does not alter any of the general conclusions.
IV. DESCRIPTIVE STATISTICS
Table 1 presents statistics describing the outcomes that we use to proxy reputation. The publication-based ranking (again, 1 is highest) has a mean well below the average of 200 departments, partly because higher ranked departments are larger, partly because U.S. economics departments disproportionately comprise the higher ranks of the worldwide set of 200 institutions. The USNWR and NRC ratings (with 5 being the maximum possible in both, and 2.5 and 0 the respective minima) indicate similarly that the average sample member's location is in a department that is fairly highly rated. Only 3% of the full professors have been Honored by the Nobel Committee or the AEA, with honors, having been received by individuals in only 18 departments. Econometric Society Fellows comprise 21% of the sample and are found in 46 of the departments.
Table 2 contains descriptive statistics on the possible determinants of reputation. The first thing to notice is the skewness of all the quality measures. The quantity measure is also highly skewed, although not nearly so much as the quality measures. In this sample 6% of the members are women, roughly consistent with a recent survey (CSWEP 2009) showing 8.5% females among full professors at a small number of Ph.D.-granting institutions. Nearly one-fourth of the sample members received their undergraduate education in a country where English is not the predominant language and we classify 9% as econometricians.
If the quality measures were nearly perfectly correlated, examining how various dimensions of quality affect reputation would be a futile exercise. They are not. Only 12 of the 20 most heavily cited members of our sample are among the top 20 in terms of the scholar's most-cited single publication. Although the correlation of these two measures is high, 0.81, there may be enough independent variation to allow us to examine the roles of both dimensions of quality in generating reputation. The correlations of Q are much lower--0.56 with [q.sub.1], 0.32 with [q.sub.21]. Only 4 of the 20 most prolific authors are in the top 20 along either of the quality dimensions. Publishing papers represents a different dimension of activity from total measured quality or the quality of one's best-known work as indicated by citation counts.
V. THE IMPACT OF QUANTITY AND QUALITIES ON REPUTATION
A. Reputation Reflected in Awards
The first three columns of Table 3 report the results of estimating the impacts of Q, [q.sub.1], and [q.sub.21] on the probability of receiving one of the rare honors available to U.S. economists. The second three columns present estimates of the determinants of having been elected a Fellow of the Econometric Society. In addition to the results displayed in the table, the probits describing Honored all hold constant for alphabetical location, location of undergraduate education, and a quadratic in the year the scholar's first SSCI-indexed paper appeared, whereas the probits for election as Fellow add the indicator for female to these. (9) Place in the alphabet never has a significant effect on receipt of one of these awards (and has no significant effect on any of the measures discussed in this section), nor does gender. Not surprisingly, being Honored is substantially more likely among authors whose first published paper appeared earlier--except for the Clark Medal, these awards are usually for a lifetime of work.
The results presented in column (4) are similar to those in column (1), as are those in column (5) to column (2), and in column (6) to column (3). The first point to note is that the number of entries, Q, never has a significant positive impact. Conditional on quality, having produced a lot of material adds nothing. Indeed, the effect on Fellow is negative and significant in all three specifications. At the very least we can conclude that quantity does not add to these proxies for academic distinction and may diminish it, conditional on the quality of the work. (10)
There is some evidence in Table 3 that both dimensions of quality matter, although the returns to [q.sub.21], conditional on [q.sub.1], are only barely statistically significant in the equations for Honored and insignificant in the equations for Fellow. There is also evidence that the marginal payoff to additional citations in total, or to the author's most-cited work, is diminishing: Adding quadratic terms to these probits, as shown in columns (2) and (5), substantially increases their ability to predict the receipt of these awards. (11)
Also intriguing in Table 3 are the changes in the estimates that occur when we recognize that, except for the Nobel Prize, each of the other honors is awarded to U.S. economists on a regular basis. Even if Q or each measure q were smaller, some U.S. economist would have his/her achievements acknowledged by receipt of one of the AEA awards; and while it is not necessary, one might imagine that current Econometric Society Fellows, of whom half are Americans, would continue to elect many of their American peers. Columns (3) and (6) are identical to columns (1) and (4), except that Q and each measure q are proxied by the scholar's rank along each dimension. Here and throughout we treat rank as increasing, so that a positive coefficient indicates that higher quality or quantity increases the likelihood of the outcome that reflects an enhanced reputation. In the case of Honored the equation does not fit as well as the quadratic version, but the fit is better than in column (1). The ability to predict election as Fellow is, however, substantially enhanced. (12) These results suggest that receipt of these awards may be more like a tournament than a competition in which additional quality per se increases the chance of success (Lazear and Rosen 1981).
B. Individual Reputation Reflected by Departmental Reputation
In this subsection, we examine how the reputation of the economics department with which an individual is affiliated is related to the quantity and quality measures that we believe may determine reputation. In order to maintain the assumption that a department's reputation proxies that of an individual member, we arbitrarily restrict the samples in this section to departments with at least ten full professors (so that presumably an individual's reputation has only a small part in establishing the reputation of the collective and problems of reverse causation are minimized). This reduces the number of observations from 1,339 to 1,188 (and the number of departments to 66). (13)
The determinants of the scholarly reputations of the scholars (proxied by the reputations of their departments) are presented in Table 4. Included in all the equations, but not shown in the table, are the effects of experience, alphabetical position, being an econometrician, and gender. None of the last three came close to statistical significance in any of the estimates. Given the importance of size, the number of full professors in a department is held constant and is unsurprisingly highly significant.
Columns (1) and (5)present results analogous to those in columns (1) and (4) of Table 3. We again treat the publication ranking so that a higher number indicates a higher rank, and thus that all the independent variables with positive coefficients indicate an improvement in the rankings. In nontabulated estimates younger authors (those whose first papers are more recent) are associated with departments with higher reputational rankings. Holding this measure constant may be important and may reflect the crucial nature of one's first publication (Siow 1991); but the result is at least partly an artifact of the sample selection criterion we have used--full professors. Those scholars who become full professors earlier tend to be associated with schools with higher reputations. (14) This outcome may result because those schools have sufficient resources to gamble on very promising younger researchers.
Having been educated in an English-speaking country (for the overwhelming majority of the sample this is the United States) is associated with being at an institution with a lower reputation. (15) This is similar to the results for election as Fellow, although having been educated abroad had no impact on being Honored by the Nobel Committee or the AEA. These results may reflect self-selection by potential scholars from the upper tail of ability among potential non-American graduate students and/or faculty. Or it may reflect the unwillingness of lower-ranked institutions to hire otherwise identical foreigners, either because the schools are more concerned about language ability or because of pure discrimination.
As with the results on awards, at the very least here too Q has no impact. The estimates in columns (1) and (5) of Table 4 show, however, that increases in [q.sub.1] lead the scholar to be located in a higher ranked department. There is no extra fillip to these measures from having one's scholarly recognition concentrated more heavily on one work--[q.sub.21] has no effect on the USNWR rating and an insignificant positive effect on the publication-based ranking. When we allow for nonconstant marginal returns to the quality measures in the estimates shown in columns (2) and (6), there are diminishing returns to quality along both dimensions. Moreover, both linear and quadratic terms along the overall quality dimension are statistically significant, but the quadratic in [q.sub.21] only matters for the publication-based ranking. Implicitly, the results demonstrate that one's reputation, as proxied by the rank of one's department, is based on the overall quality of one's publications, as measured by the total recognition by other scholars, and possibly too by the distinction of one's best-known work. It is unaffected by the quantity of publications.
Unlike the results in the previous subsection, there is no mechanical reason to expect any relation between a scholar's rank along some quality dimension and the ranking of the department with which she/he is affiliated. Departments and universities are, however, competing for prestige/students/funds, so that at least to some extent one might imagine that there are tournament-like aspects to the market for individuals' reputations, as reflected in the rankings of their departments. To examine this possibility we reestimated the basic equation, again proxying Q, [q.sub.1]. and [q.sub.21] by the scholar's rank in the sample along the criteria of number of publications, total citations, and citations to his/her most-cited work.
Comparing the results presented in columns (3) and (7) to those in columns (1) and (2), and (5) and (6), we see that a scholar's rank along all three dimensions has a bigger impact on reputation than do the cardinal measures. The results suggest that the market for scholarly reputation has tournament-like characteristics along one dimension of quality. In this formulation, however, [q.sub.21] is not important--only total citations matter. Moreover, the impact of one's rank in Q on both measures is significantly negative.
One might be concerned about a possible two-tier labor market in economics, with the determination of reputation differing between public and private institutions. That argument is especially cogent given that the top eight departments in both the publication-based rankings and the USNWR ratings are in private institutions. In fact, there is very little difference between the two types of institution. Separate estimates of column (3) for public (private) schools yield estimates of the effects of [q.sub.1], [q.sub.21], and Q as 6.264 (6.080),--7.806 (-1.237), and -2.995 (-1.796), respectively. Similar reestimates of column (7) yield 0.564 (1.036), -0.0105 (-0.0096), and -0.0071 (-0.0416). Tests of equality of the estimates across the equations fail to reject the hypothesis that the structures are the same. This similarity should not really be surprising--there is substantial mobility between tenured positions across the two sectors, so that reputational effects are likely to be determined at the same margins.
We can expand the quality ranking measures to use both [q.sub.21] and [q.sub.22], as shown in columns (4) and (8) of Table 4. (16) As in columns (3) and (7), only the scholar's ranking in total citations has a significant effect on the proxies for reputation. There is no evidence of a significant additional impact of either the most- or second-most-cited paper. What matters is the overall quality of an author's works. (17)
Do the effects on reputation differ at the margin depending on the scholar's ability to generate reputation at different levels? Quantile regressions at the 75th percentile, the median, and the 25th percentile of the quality rankings of departments suggest that there are few differences in the effects of the significant variables, [q.sub.1] and Q, at these quantiles. The importance of total quality and the negative impact of quantity pervade the distribution. The same conclusions are produced from separate least-squares regressions on individuals in the upper and lower halves of the publication-based rankings and the USNWR ratings.
While the effects of total quality and quantity have statistically significant (in opposite directions) effects on these measures, the sizes of the effects are at least as important. Using the distributions in Table 2 and the estimates in columns (3) and (7) of Table 4, we can calculate the impacts of changes in these measures. Taking the publication rankings, a move from the 25th to the 75th (5th to 95th) percentile of [q.sub.1] increases the ranking by 42 (75). Using the USNWR ratings, the same changes generate effects of 0.62 (1.11) on the rating. The inter-quartile change thus increases the reputational proxies by about two-thirds of a standard deviation, while a change from the 5th to the 95th percentile would move one's reputation from near the middle to near the top of the range. Increases in Q significantly reduce the reputational proxies, although the effects are smaller but still substantial: -18 (-32) in the publication ranking, -0.15 (-0.27) in the USNWR rating.
C. Using Mobility to Bound Reputation
An additional way of examining the roles of quantity and quality is to study the reputation of the department into which new full professors were hired between 1993 and 2007. We continue to assume that the decision to hire a new full professor, either from outside or by granting tenure to an untenured professor, reflects decision-makers' beliefs that the person's reputation is at least that of the collective's average. Accordingly, a lower-bound measure of the individual's reputation is the reputation of the department some time shortly before she/he joined it.
For the 308 scholars in our sample who moved between schools from 1993 to 2007 or who entered the economics labor market after 1992, we relate the NRC93 quality rating to the variables included in the equations presented in Table 4. The scholarship of these 308 individuals could not have affected the quality rating of the school in 1993. Thus relating the NRC93 rating to the individuals' efforts is an even cleaner measure of the impact of individuals' quantity and quality inputs than those used in the previous subsection, under the maintained assumption that schools will not hire or tenure scholars who they believe will reduce their ranking.
The estimates are shown in Table 5. As in all the other estimates in this section, rankings of the quality and quantity inputs describe the determinants of reputation better than cardinal measures. Also as before, there is a negative relationship between this proxy and the scholar's ranking in the distribution of Q. Again, merely writing more papers, conditional on the overall quality of one's work, generally reduces one's reputation. As the results in columns (1)-(3) show, only [q.sub.1] significantly increases what we have identified as the lower bound of the scholar's reputation. Moreover, as in nearly all the other equations, the marginal effect of additional quality is diminishing, and, as before, increasing [q.sub.21] has essentially no effect on reputation. (18) Finally, as with the larger sample in Table 4, we cannot reject the hypothesis that the same equation characterizes behavior in both public and private institutions.
The specification using rankings is an arbitrary transformation of the underlying measures. Perhaps our conclusions about the importance of rank in these hierarchies are not robust to tests of alternative specifications. We experimented with several of these in the estimates in Tables 3-5. In no case did a set of cubics in the quantity measure and the two measures of quality add significantly to the explanatory powers of the various equations. Similarly, using the inverses of each of the three measures reduced the equations' explanatory power. Logarithmic transformations improved the explanatory ability in some cases, but reduced it in others. Like rank measures, logarithmic transformations reduce the variance of these highly skewed variables. Thus, it is unsurprising that it is difficult to distinguish between them and the ranks themselves. Seen in this light, these experiments can be viewed as providing additional support for the role of rankings in determining these reputational rewards.
VI. MONETARY RETURNS TO QUALITY AND QUANTITY--AND A CONUNDRUM
A long literature exists on the roles of the quantity and quality of publications in salary determination in economics, with citations by other scholars often being the proxy for quality and with counts of articles and books proxying quantity. (Bratsberg, Ragan, and Warren 2010; Hamermesh, Johnson, and Weisbrod 1982; and Moore, Newman, and Turnbull 1998 are just a few of the many that typically use samples of economists at a few institutions. A plethora of similar studies exists for other disciplines too--e.g., Long, Allison, and McGinnis, 1993, in sociology.) Some of these studies suggest that quality is more important, others that quantity of publications matters too. This literature is distinct from the approach used thus far here; but since we assume that economics departments and the universities of which they are constituents attempt to maximize reputation, one might expect that the inputs that affect reputation will affect salaries in similar ways. The specific question is whether for this large sample the determinants of salaries and of the proxies for reputation are the same.
A. The Effects of Quality and Quantity of Research on Salaries
Salary data for individual faculty members are difficult to obtain (nearly impossible for private institutions), but for 43 of the 53 public institutions in our sample, including the ten highest-rated public-university departments, we acquired data from university websites and direct contacts to calculate full-time academic-year salaries. (19) After various consistency checks we were left with usable observations on the academic-year salaries of 564 economists. (20) Although we only have data on 42% of the sample, this group should be sufficiently large to allow us to explore a comparison of the impacts of Q, [q.sub.1], and [q.sub.21] on salary to their impacts on reputation.
The observations in this subsection are not a random subsample: Unsurprisingly, as public universities are typically lower-ranked, sub-sample members have fewer citations in total (median of 250 compared to 310 in Table 2), fewer citations to their most-cited paper (median of 57 compared to 71), but nearly the same number of papers (median of 22 compared to 24). (21) Moreover, those individuals on whom we have usable salary and citations data have statistically insignificantly lower averages of Q, [q.sub.1], and [q.sub.2n] than those public-university faculty members without salary information, mainly because we could not calculate 9-month salaries for a few productive scholars in administrative positions.
Regressions of the logarithms of 9-month salaries on the same combinations of citations and publications measures used in the previous sections are shown in columns (1)-(3)of Table 6. The usual control variables are included too, along with school fixed effects. (22) As with the reputational measures, salaries are described better by ranks than by raw numbers* Thus, we concentrate on the results in column (3). (23) As before, total citations have a significant positive effect on salary. The effect is not small: Going from the 25th to the 75th percentile of total citations in this subsample raises salary by 33%, and from the 5th to the 95th by 67% (on a mean of $150,200).
Unlike its impact on reputation, which for some of the proxies was positive, there is no significant additional effect of citations to the author's most-cited paper--indeed, it seems to have less of an effect on salary than citations to the scholar's other work. Moreover, unlike the results on direct measures of reputation, but as shown in some studies of academic salaries, Q has a positive and statistically significant effect on salary. The effect is moderate: going from the 25th to the 75th (5th to the 95th) percentile of Q increases salary by 6 (10%), but it stands in sharp contrast to the results on the proxies for reputation. Estimates of these effects at the 25th, median, and 75th percentiles (not presented) show that the effects are quite similar at different points of the salary distribution.
One might be concerned that these results arise simply because the subsample differs along quality dimensions from the overall sample. To examine this possibility, in columns (4)-(6) of Table 6 we reestimate the equations in column (6) of Table 3 and columns (3) and (7) of Table 4, using only the subsample of public-university faculty on whom we have salary data. (24) The impacts of Q and [q.sub.1] on Fellow are smaller in absolute value in this subsample, but they are qualitatively not that different from those for the entire sample. The estimated impacts of the quality and quantity measures on departmental rankings and ratings are nearly identical. The results suggest that the effect of Q on an individual's salary is at the very least quite different from its effects on a large number of proxies for reputation. (25)
B. Why the Difference Between Reputation and Salary?
Although some of their determinants are the same, clearly reputation does not translate directly to salary nor can they be equated. The contrast between the effects of Q in column (3) and in columns (4)-(6) thus poses an interesting conundrum: Why should universities pay off on something--the sheer volume of production--that, as our results demonstrate, does not raise and may even reduce an individual's reputation? There are two general approaches to explaining away this remarkable difference: (1) The information available to those who determine salaries differs from that available to those who determine reputation and (2) there are unobservable factors correlated with Q that make scholars who produce a lot of output more productive for their employers in ways that do not raise and might lower their external reputations.
We can test various explanations under the first general rubric. One possibility is that those who determine salaries are unaware of citation information or do not take it into account, so that Q provides the only signal of productivity (remember that Q and the q measures are positively correlated). As one effort to examine this possibility, we surveyed individuals in the departments used in this subsection, obtaining data on whether information on recent accepted/published work, and/or recent citations, is collected for use in determining annual salary changes. (26) All 43 departments obtain information on publications, but only eight departments, covering 126 of the 564 individuals included in the estimates in Table 6, obtain information on annual citations. The contrast between the results on salary and reputation may therefore arise because most schools ignore citation information in salary setting.
If so, we would expect to see that the number of publications affects salary determination less in those departments that do collect citations data, while total citations and citations to the most-cited paper have greater effects there. Adding interactions of an indicator for collecting citations data with Q, [q.sub.1], and [q.sub.21] to column (3) of Table 6, as shown in column (1) of Table 7, we find that none of the interactions has a t statistic above 1, and the vector of interactions is statistically insignificant. Entering each interaction term separately does not alter the conclusion. Although some schools do collect citation data when salaries are determined, having that information available does not seem to alter salary determination.
A related possibility is that the locus of control of salary setting may differ across the 43 institutions, with salary setting in departments that are part of relatively homogeneous colleges paying less attention to Q, as other departments in the unit have similar production technologies and internal systems of evaluation. To examine this possibility we created an indicator for those departments that are part of large general colleges (typically called Arts and Sciences, Liberal Arts, etc.), equaling 1 for 41% of the observations included in Table 6. The results of interacting this indicator with the quality and quantity measures are shown in column (2) of Table 7. Again, none of the interactions has a t statistic above l and the entire vector is insignificant. (27)
Another explanation may be that the salaries of those who have moved into a department are determined differently from those of stayers, with the determination of movers' salaries being more responsive to Q and less to the quality measures. For the reduced sample who could have moved from another institution, we interact the indicator of movement with the quality and quantity measures, with the results presented in column (3)of Table 7. Again none of the individual interactions is significant nor is the entire vector of interactions as a group. (28)
Yet another possibility under this rubric is that salary determination in larger departments, where there is a greater likelihood that a colleague is familiar with the quality of one's work and does not need to rely so much on quantity as a signal, is based less heavily on Q. Interacting the number of full professors in a department with Q and the two quality indicators, as shown in column (4)of Table 7, produces the same insignificant interactions as the previous three respecifications.
One (perhaps testable) possibility under the second rubric is that the same personality characteristics that lead scholars to churn out many papers are correlated with other characteristics that generate high salaries. Some of the correlations may be with characteristics that might be viewed as desirable. For example, those who write a lot might also be energetic or even better teachers, might be more active departmental citizens, and/or less troublesome and more stable colleagues. Some might be viewed as undesirable--they might agitate more successfully for salary increases, either by soliciting and receiving job offers and/or by clever negotiation with their current employer.
To examine this possibility, admittedly quite imperfectly, we obtained (for all but two people) information on whether individuals in the subsample had ever been a department chair or a more senior university official. Presumably being chosen for such a position indicates at least a minimal level of organizational ability and desirable personality traits. We interact an indicator for this proxy for personality traits with the quantity and quality measures, with the results shown in column (5) of Table 7. (The indicator equaled 1 for 24% of the subsample.) As a group the interactions are statistically significant, and [q.sub.1] matters relatively more than Q for those who were never administrators, which gives some credence to this explanation for the disconnect between reputation and salary. Nonetheless, even for those who never were administrators, Q has a positive and nearly significant effect on salary.
Some additional, nontestable explanations combine both informational inadequacy and unobservables. Assume that entrants into the academic market for reputation consist of two observably indistinguishable but distinct types: good guys, whose future papers will be cited a lot; and opportunistic guys, who devote their effort to publishing articles that eventually generate few citations. Reputation alone cannot fully sort the good from the opportunistic, because updating wages may not be fully possible when wages are not adjustable downward. In the academic market, a disconnect between reputation and salary will occur. As another possibility assume that each supplier of reputation produces along a stochastic Leontief production function, so that each paper has the same probability of being an important contribution, but some are of higher quality than others. Then greater quantity, indicative of more effort (more draws from the person's distribution), indicates that the scholar is more likely to produce a high-quality piece of research. Quantity itself becomes an indicator of future quality. Clearly this explanation too is not testable, but it might be an additional explanation for the findings that the quantity of scholarly output per se has negative effects on reputation but positive effects on salary.
The difficulty with both of these observations, however, is that over 60% of the observations in our sample were over age 50, long past the point at which most economists have produced what will eventually be their best-known works. (29) In the end, none of the testable explanations explains the conundrum very well. It is a very interesting topic for future research, both within the economics profession and, mutatis mutandis, in academe more generally.
Reputation is a nebulous concept, so we have used a number of proxies for it, none of which alone could fully capture the idea, to examine its determinants among academic economists. Our focus has been on the relative roles of what might be characterized as the number of attempts to establish reputation, the number of actual impressions made on those who might determine one's reputation, and the distribution of those impressions across their number. We proxy the number of attempts by the number of papers a scholar has published, their impressions by the number of times those works are cited, and their distribution by the concentration of citations on one or several papers.
Although the evidence is somewhat mixed, in this example simply writing more papers has no impact on a large number of proxies for reputation and very possibly even a negative effect. The major determinant of reputation--what is rewarded in this particular academic reputational market--is the interest that a scholar's work generates among his/her peers. There is at most only weak evidence that the concentration of impressions on a single piece of work--one or two articles, in this case--generates additional increases in one's reputation. Reputation in this profession has tournament-like aspects--one's ranking along the dimension of overall quality appears to describe proxies for reputation better than do absolute achievements along this dimension.
In contrast to their effect on reputation, the number of publications alone has significant positive effects on salary. Together with the results on reputation, these results suggest that generating the respect that influences the direction of a field, and thus scientific progress, comes from creating works that are viewed as important. Merely publishing papers does not make one, to use the popular term, an "opinion leader," although it is privately monetarily productive. Similarly, being a "one-hit wonder" generally does less for one's contribution to scientific progress than a stream of papers of high quality.
Although our example is specific, the general view of a market in which various characteristics interact to generate reputation seems useful in a variety of other markets. Any labor market where participants' output can be identified would appear equally amenable to this approach. Other academic disciplines and the professions (attorneys and physicians being obvious examples for which overall measures of reputation are readily accessible) could be analyzed using similar methods; and the nature of reputation in artistic/creative activities is similar enough to that in academic disciplines to make studying it using approaches like the one here worth pursuing. Also, the reputations of restaurants, and perhaps other retail outlets, might be studied in the same way. The main points are that it is useful to view the establishment of reputation as stemming from the quality and, although not in economics, the quantity of what we produce, and from the preferences of those people who determine our reputations.
AEA: American Economic Association
Arnold, M. M., and D. S. Lindsay. "Remembering Remembering." Journal of Experimental Psychology: Learning Memory and Cognition, 28, 2002, 512-29.
Basu, K. "A Theory of Association: Social Status, Prices and Markets." Oxford Economic Papers, 41, 1989, 653-71.
Blank, R. "The Effects of Double-Blind Versus Single-Blind Reviewing: Experimental Evidence from the American Economic Review." American Economic Review, 81, 1991, 1041-67.
Borjas, G. "Self-Selection and the Earnings of Immigrants." American Economic Review, 77, 1987, 531-53.
Bratsberg, B., J. Ragan, and J. Warren. "Does Raiding Explain the Negative Returns to Faculty Seniority?" Economic Inquiry, 48, 2010, 704-21.
Cole, J., and S. Cole. Social Stratification in Science. Chicago: University of Chicago Press, 1973.
CSWEP. Newsletter, Winter 2009.
Einav, L., and L. Yariv. "What's in a Surname? The Effects of Surname Initials on Academic Success." Journal of Economic Perspectives, 20(1), 2006, 175-88.
Feist, G. J. "Quantity, Quality and Depth of Research as Influences on Scientific Eminence: Is Quantity Most Important?" Creativity Research Journal, 10, 1997, 325-35.
Ferber, M. A. "Citations: Are They an Objective Measure of Scholarly Merit?" Signs, 11, 1986, 381-89.
Hamermesh, D. S. "Facts and Myths about Refereeing." Journal of Economic Perspectives, 8, 1994, 153-63.
--. "Doing Applied Economics: Normative and Positive Aspects," in How Do Economists Do Economics? edited by S. G. Medema, and W. J. Samuels. Cheltenham: Elgar, 1996.
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.
Hamermesh, D. S., and P. J. Schmidt. "The Determinants of Econometric Society Fellows Elections." Econometrica, 71, 2003, 399-407.
Hayes, S. C. "When More Is Less: Quantity Versus Quality of Publications in the Evaluation of Academic Vitae." American Psychologist, 38, 1983, 1398-1400.
Hirsch, J. E. 2005. "An Index to Quantify an Individual's Scientific Research Output." Proceedings of the National Academy of Sciences, 102, 2005, 16569-72.
Kalaitzidakis, P., T. P. Mamuneas, and T. Stengos. "Rankings of Academic Journals and Institutions in Economics." Journal of the European Economic Association, 1, 2003, 1346-66.
Kurtz, K., and J. Loewenstein. "Converging on a New Role for Analogy in Problem Solving and Retrieval: When Two Problems Are Better Than One." Memory and Cognition, 35, 2007, 334-41.
Lazear, E., and S. Rosen. "Rank-Order Tournaments as Optimum Labor Contracts." Journal of Political Economy, 89, 1981, 841-64.
Long, J.S., P.D. Allison, and R. McGinnis. "Rank Advancement in Academic Careers; Sex Differences and the Effects of Productivity." American Sociological Review, 58, 1993, 703-22.
Mailath, G., and L. Samuelson. Repeated Games and Reputation: Long-Run Relationships. New York: Oxford University Press, 2006.
Moore, W. J., R. J. Newman, and G. K. Turnbull. "Do Academic Salaries Decline with Seniority?" Journal of Labor Economics, 16, 1998. 352-66.
National Research Council. Research-Doctorate Programs in the United States; Continuity and Change. Washington. DC: National Academy of Sciences, 1995.
Ransom. M. "Seniority and Monopsony in the Academic Labor Market." American Economic Review, 83, 1993, 221-33.
Rosen, S. "Some Economics of Teaching." Journal of Labor Economics, 5, 1987, 561-75.
Slow, A. "Are First Impressions Important in Academia?" Journal of Human Resources, 26, 1991, 236-55.
Starns, J. J., and J. L. Hicks. "Source Dimensions Are Retrieved Independently in Multidimensional Monitoring Tasks." Journal of Experimental Psychology: Learning, Memory, and Cognition, 31, 2005, 1213-20.
Tadelis, S. "The Market for Reputations as an Incentive Mechanism." Journal of Political Economy, 110, 2002, 854-82.
Tulving, E., and D. M. Thomson. "Encoding Specificity and Retrieval Processes in Episodic Memory." Psychological Review, 80, 1973, 352-73.
Ward, B. "The Firm in Illyria: Market Syndicalism." American Economic Review. 48, 1958, 566-89.
(1.) Merriam-Webster Online Dictionary, June 5, 2008.
(2.) Obviously, we could generalize this to additional dimensions of quality, but the formulation we use is the simplest that allows us to construct empirical analogues.
(3.) The minimum size of the collective whose reputation can be assigned to the individual for this purpose and is thus not determined by that individual is not clear, and in the empirical work we experiment with various cutoffs.
(4.) The 88 departments range from Harvard to the University of Arkansas, Fayetteville.
(5.) Most department administrators are admonished to build a better department, one that is stronger in research. Thus, for example, one university describes the chair's role, "The chairperson has a special obligation to build a department strong in scholarship...." (Michigan State University, Faculty Handbook, 22.214.171.124).
(6.) One might be concerned that excluding publications that are in science journals, and also excluding citations from science journals, might bias our results. It is obviously true that this exclusion generates errors; but the errors are clearly small--in a random sample of 50 observations (chosen from a consecutive section of the alphabet) the correlation between citations in the SSCI and total citations in the SSCI, Science Citation and Arts and Humanities Indexes together was 0.979, while that between citations to the most-cited paper from the two sources was 0.993. One might also be worried that we count all citations, including self-citations. To examine this potential problem we took another random sample of 50 and computed the correlation of total citations and total citations without self-citations. The Pearsonian correlation coefficient was 0.984, while the rank correlation coefficient was 0.991. We thus do not view this as a problem. (In any case, while it is easy to obtain total citations excluding self-citations, obtaining [q.sub.2n] excluding self-citations is not feasible.)
(7.) This choice seems the best among the possible ways of counting total citations and citations to individual works. One should stress, however, that Q, and thus the publications that could be cited, excludes books and working papers. The former exclusion is not important for most economists, and the latter exclusion matters little in a sample of full professors. The alternative use of the SSCI would be based on authors rather than publications, but the SSCI does not allow a convenient tabulation of a scholar's most-cited works by this method. An alternative would be to use Google Scholar or SCOPUS, but their methods of tabulation are less clear.
(8.) The results of Cole and Cole (1973) make it clear that concerns that citations might measure infamy rather than fame are misplaced.
(9.) The indicator for female is not included in these probits. because only one woman had been honored according to this measure.
(10.) One might be concerned that we measure from the stock of existing Fellows, not from the flow of newly elected Fellows. To do that we would require information on Fellows at the time they were elected, which is difficult but possible. Fatally, however, it would require information on those who were never elected, most of whom were never on the Fellows ballot. While using the stock may introduce errors, there is no reason for them to be generating biases that lead these results to mirror those for all the other measures of reputation discussed in this section.
(11.) We do not use the h index (Hirsch 2005). (Ranking an author's papers in descending order of their citation counts, an author's h value indexes the paper that is ranked hth in the order and that receives h citations. An h index of around 35 is typical for Nobel laureates in economics.) For analytical purposes this measure has the problems that it combines quantity and quality and also fails to indicate dispersion in quality. No doubt it could be used here, but interpreting the meaning of any measured impact would be difficult.
(12.) The equations in columns (1)-(3) were reestimated defining the outcome to exclude receipt of the Clark Medal. The results were very similar to those presented in the table. Similarly, none of the conclusions here or anywhere else in this study changes qualitatively if we estimate bivariate relationships between a reputational outcome and each of the quality and quantity measures singly.
(13.) The crucial results in this subsection hardly change if we restrict the subsample to the 940 scholars located in the 44 departments with at least 15 full professors, or if we include all 1.339 observations.
(14.) Dropping this measure from the regression hardly changes the results: The coefficient on [q.sub.1] becomes 0.511. that on [q.sub.21] becomes 0.945. and that on Q becomes -4.882.
(15.) There may be a distinction between those educated on the Indian subcontinent and those educated elsewhere, perhaps due to differential discrimination, perhaps to differential familiarity with English. We added an indicator for Indians and Pakistanis. but its effect was tiny and statistically insignificant.
(16.) The simple correlation between the numerical proxies for these quality measures is 0.82.
(17.) The partial correlation coefficient between [q.sub.1] and the vector [q.sub.21] through [q.sub.25] is 0.974, making it difficult to move beyond the estimates presented in column (4).
(18.) One might reasonably argue that the effects of Q, [q.sub.1]. and [q.sub.21] on this proxy for reputation differ between faculty who moved between tenured positions and those who first achieved tenure. To examine this possibility the models in Table 5 were reestimated separately for the 130 newly tenured faculty and the 178 faculty who moved. While the positive effect of quality and negative effect of quantity were larger among the newly tenured, we cannot reject the hypothesis that the structure of the relationship is the same for both groups.
(19.) Most of the data were for 2007-2008, but where they were not we used [1.04.sup.(2007-t)] to inflate salaries. One institution only had salary data for 2004-2005, while for many, particularly the University of California system, data were for 2006-2007. In some cases, it was impossible to determine if the faculty member was on a 9-month appointment, and those cases were not included in the analysis.
(20.) Sixteen of the schools included here comprise the biennial survey of salaries reported for 2006-2007 at http://www.eco.utexas.edu/faculty/Hamermesh/EcSalsPublic Cleaned.xls. The correlation of the average salaries computed for each school and the averages provided there is 0.89, suggesting relatively little systematic measurement error in our compilation of salary data.
(21.) Indeed, the distributions of citations and numbers of publications are similarly below the median, but the overall distributions have longer fight tails.
(22.) While the vector of fixed effects is highly significant statistically, its inclusion changes none of the inferences about the impacts of the citations and publication measures on salary. The results are also qualitatively nearly identical if we replace the school fixed effects with the departmental rankings, or the departmental ratings, used in the previous section: and the correlation of ranking (rating) with average salary is 0.80 (0.84).
(23.) As in Tables 3 and 4 we experimented with several alternative specifications of Q, [q.sub.1] and [q.sub.21]. As in those tests, the logarithmic specification explained the variance of the outcomes about as well as the ranks, while the inverse and cubic performed substantially worse.
(24.) We do not use Honored here, as only 5 of the 564 observations received an honor, while 63 are Fellows.
(25.) These results are not a matter of somehow different salary determination for younger and older full professors. Reestimating the specification in column (3) on the large majority of the sample who received their doctorate before 1987 hardly changes the results. Similarly. interacting experience with the quantity and quality measures adds nothing to the estimates in column (3). Nor are salaries lower, other things equal, as a compensating differential for those who have better reputations: For those scholars for whom Fellow = 1, in reestimates of column (3) salaries are a significant 10% higher, other things equal; but the qualitative impacts of Q, [q.sub.1], and [q.sub.21] are unchanged.
(26.) The e-mail questionnaire was: "In doing annual merit/salary reviews, what information is requested from faculty members in your Department? (1) List of articles accepted and/or published during the year--YES or NO. (2) List or count of citations during the year to published or unpublished work--YES or NO. Please delete the incorrect answer to each of these two questions."
(27.) One might also guess that higher-ranked schools pay more attention to total citations, which, as we showed, are the major determinant of reputation, and less to Q. Reestimating the equation in column (3) of Table 7 by adding the interactions of the department's ranking (rating) with Q, [q.sub.1], and [q.sub.21] suggests that this is not the case in this subsample. The interaction terms were insignificant individually and as a group, F(3, 550) = 0.92, (F(3, 348) = 0.64).
(28.) The estimates imply that, other things equal, those who had moved earned 14% more per year than otherwise identical faculty who had not. This result is consistent with the Ransom's (1993) findings on the relation between academic salaries and job tenure.
(29.) Only 10% of the authors of articles in the top three general journals in economics in 2003 were over age 50; and even that percentage far exceeded the representation of these "elderly" scholars in articles published in these journals in earlier years (Hamermesh 1996).
DANIEL S. HAMERMESH and GERARD A. PFANN *
* We thank Randall Akee, Jason Faberman, Mark Harrison, Ken Hendricks, Andrea Ichino, Lawrence Kahn, Arthur Markman, Gerald Oettinger, James Pennebaker, the late James Ragan, Joe Stone, Mark Walker, two referees, and participants in seminars at several universities and IZA. Karen Mulligan and Amanda Smith provided very careful research assistance.
Hamermesh: Sue Killam Professor in the Foundations of Economics, University of Texas at Austin, and Professor of Labor Economics, Maastricht University, IZA and NBER. Phone 001 512 475-8526, Fax 001 512 4713510, E-mail firstname.lastname@example.org
Pfann: Professor of Econometrics and Organization, Maastricht University, CEPR and IZA. Phone 0031 43 388-3832, Fax 0031 43 388-4856. E-mail g.pfann@ maastrichtuniversity.n
TABLE 1 Means and Standard Deviations, Outcome Measures, Full Professors in Top-Rated Departments, 2007-2008, N = 1,339 Except Where Otherwise Noted Outcome Publication-based ranking (1 top, 200 bottom) (a) 62.92 (57.20) USNWR rating (5 highest) 3.88 (N = 865) (0.78) NRC93 faculty rating (5 highest) 3.25 (N = 1,179) (1.02) Moved 1993-2007 0.197 (N = 913 eligibles) Honored 0.031 Honored (w/o Clark) 0.025 Fellow 0.207 Note: (a) From Kalaitzidakis et al. (2003). TABLE 2 Descriptive Statistics, Personal Measures, Full Professors in Top-Rated Departments, 2007-2008, N = 1,339 Percentile Input Mean Minimum 5 25 50 Citations: [q.sub.1] 685 0 25 131 310 [q.sub.2] First paper 150 0 9 32 71 Second paper 82 0 5 21 44 Third paper 59 0 3 15 32 Fourth paper 46 0 2 11 25 Fifth paper 38 0 1 8 20 Q 31.76 1 7 14 24 Female 0.060 -- -- -- -- English-language education 0.778 -- -- -- -- Econometrician 0.087 -- -- -- -- No. of full professors 19.45 3 7 13 17 Percentile Input 75 95 Maximum Citations: [q.sub.1] 727 2,575 14,232 [q.sub.2] First paper 155 498 4,580 Second paper 91 260 2,212 Third paper 68 189 1,059 Fourth paper 53 150 879 Fifth paper 42 124 717 Q 39 79 283 Female -- -- -- English-language education -- -- -- Econometrician -- -- -- No. of full professors 24 39 39 TABLE 3 Determinants of Various Honors, Probit Derivatives, N = 1,339 Honored Ind. Var.: (1) (2) (3) Total citations/1,000 0.0043 0.0042 -- (3.57) (2.46) (Total citations/ 1,000) (2) -- -0.00026 -- (2.63) Citations to most-cited paper/1,000 0.0103 0.0099 -- (2.14) (1.84) (Citations to most-cited paper/ -- -0.0015 -- 1,000) (2) (1.37) No. of entries/1,000 0.015 0.0387 -- (0.37) (0.63) (No. of entries/1,000) -- -0.0131 -- (0.53) Total citations rank/1,000 -- -- 0.0034 (2.74) Citations to most-cited paper rank -- -- 0.00061 /1,000 (0.79) No. of entries rank/1,000 -- -- 0.00016 (0.30) English education 0.0028 0.0018 0.00002 (0.75) (1.03) (1.28) Econometrician -0.0071 -0.0029 -0.00002 (2.20) (2.19) (0.88) Pseudo-[R.sup.2] 0.482 0.542 0.541 Fellow Ind. Var.: (4) (5) (6) Total citations/1,000 0.296 0.351 -- (8.47) (8.65) (Total citations/ 1,000) (2) -- -0.0196 -- (6.47) Citations to most-cited paper/1,000 0.138 0.088 -- (1.08) (0.64) (Citations to most-cited paper/ -- -0.0537 -- 1,000) (2) (1.50) No. of entries/1,000 -1.861 -1.238 -- (2.70) (0.98) (No. of entries/1,000) -- -5.33 -- (0.78) Total citations rank/1,000 -- -- 0.585 (7.42) Citations to most-cited paper rank -- -- -0.0323 /1,000 (0.54) No. of entries rank/1,000 -- -- -0.124 (3.17) English education -0.200 -0.193 -0.156 (5.87) (5.97) (6.03) Econometrician 0.109 0.099 0.082 (2.38) (2.33) (2.73) Pseudo-[R.sup.2] 0.337 0.355 0.365 Note: Absolute values of t statistics in parentheses here and in Tables 4, 6, and 7, based on robust standard errors. Also included in all the probits for Honored are rank in the alphabet, location of undergraduate education, and a quadratic in the year of first paper. The probits for Fellow add an indicator for female. TABLE 4 Determinants of Measures of Departmental Reputation Publication-Based Ranking (a) Ind. Var.: (1) (2) (3) (4) Total citations/100 0.559 1.868 -- -- (1.91) (2.51) (Total citations/100) (2) -- -0.0141 -- -- (2.57) Citations to most-cited 0.946 1.773 -- -- paper/100 (1.07) (1.64) (Citations to most-cited -- -0.0439 -- -- paper/100) (2) (2.06) No. of entries/100 -5.153 -1.832 -- -- (0.78) (0.92) (No. of entries/100) (2) -- 3.980 -- -- (0.49) Total citations rank/100 -- -- 6.234 4.985 (4.36) (3.34) Citations to most-cited -- -- -2.254 -4.912 paper rank/100 (0.28) (0.64) Citations to second -- -- -- 1.432 most-cited paper (1.23) rank/100 No. of entries rank/100 -- -- -2.653 -2.446 (3.08) (3.08) English education -4.970 -7.383 -5.338 -5.253 (1.41) (1.98) (1.58) (1.53) No. of full professors 2.916 2.787 2.387 2.384 (5.55) (5.26) (4.75) (4.74) [R.sup.2] 0.331 0.359 0.421 0.422 N -- 1,188 -- -- USNWR Rating Ind. Var.: (5) (6) (7) (8) Total citations/100 0.0124 0.0391 -- -- (2.59) (4.62) (Total citations/100) (2) -- -0.00025 -- -- (4.60) Citations to most-cited 0.00034 -0.0104 -- -- paper/100 (0.02) (0.53) (Citations to most-cited -- -0.00005 -- -- paper/100) (2) (0.16) No. of entries/100 0.0164 -0.2339 -- -- (0.17) (-1.30) (No. of entries/100) (2) -- -0.0598 -- -- (0.80) Total citations rank/100 -- -- 0.0921 0.0796 (3.62) (2.39) Citations to most-cited -- -- -0.0108 -0.0137 paper rank/100 (0.77) (1.04) Citations to second -- -- -- -0.0147 most-cited paper (0.69) rank/100 No. of entries rank/100 -- -- -0.0223 -0.0202 (1.99) (1.75) English education 0.0566 0.0484 0.0516 0.0531 (1.12) (0.97) (1.13) (1.15) No. of full professors 0.0604 0.0555 0.0552 0.0552 (9.04) (8.64) (8.42) (8.42) [R.sup.2] 0.530 0.561 0.577 0.577 N -- 858 -- -- Note: The samples are restricted to individuals in departments with ten or more full professors. Also included are rank in the alphabet, indicators for female and econometrician, and a quadratic in year of first paper. Standard errors are clustered on departments. (a) From Kalaitzidakis et al. (2003). TABLE 5 Determinants of NRC93 Faculty Rating, Movers, and Newly Tenured, N = 308 Ind. Var.: (1) (2) (3) Total citations/1,000 0.199 0.541 -- (1.82) (3.37) (Total citations/1,000) (2) -- -0.0220 -- (2.42) Citations to most-cited -0.0925 0.0294 -- paper/1,000 (0.21) (0.04) (Citations to most-cited -- -0.737 -- paper/1.000) (2) (1.77) No. of entries/100 -0.194 -0.0673 -- (0.78) (0.14) (No. of entries/100) (2) -- -0.216 -- (1.24) Total citations rank/100 -- -- 0.167 (3.38) Citations to most-cited -- -- -0.0429 paper rank/100 (1.40) No. of entries rank/100 -- -- -0.0678 (2.57) English education 0.221 0.189 0.150 (2.42) (2.24) (1.86) [R.sup.2] 0.491 0.520 0.533 Note: Also included here are rank in the alphabet, indicators for female and econometrician, a quadratic in year of first paper, and the number of full professors in the department. TABLE 6 Determinants of Salary, ES Fellow, and Departmental Rankings, 43 Public-University Economics Departments Ln(Salary) Dependent Variable: (1) (2) (3) Ind. Var. Total citations/100 0.0208 0.0338 -- (6.37) (5.90) (Total citations/ 100) (2) -0.00046 -- (3.57) Citations to most-cited paper/100 -0.0275 -0.0280 -- (3.80) (2.11) (Citations to most-cited paper/ -- 0.00083 -- 100) (2) (2.14) No. of entries/100 0.191 0.4178 -- (3.74) (4.32) (No. of entries/ 100) (2) -- -0.1790 -- (3.42) Total citations rank/100 -- -- 0.0428 (5.60) Citations to most-cited paper -- -- -0.0127 rank/l00 (2.09) No. of entries rank/100 -- -- 0.0081 (2.16) p value on F test of 42 school <.001 <.001 <.001 fixed effects Adj. [R.sup.2] (Pseudo-[R.sup.2] 0.495 0.524 0.534 in column ) N 564 564 564 Fellow Publication USNWR Ranking Rating Dependent Variable: (4) (5) (6) Ind. Var. Total citations/100 -- -- -- (Total citations/ 100) (2) -- -- -- Citations to most-cited paper/100 -- -- -- (Citations to most-cited paper/ -- -- -- 100) (2) No. of entries/100 -- -- -- (No. of entries/ 100) (2) -- -- -- Total citations rank/100 0.0223 6.597 0.0665 (3.33) (3.55) (2.98) Citations to most-cited paper 0.0023 -0.429 -0.0162 rank/l00 (0.46) (0.35) (1.16) No. of entries rank/100 -0.0050 -3.151 -0.0074 (1.45) (2.56) (0.89) p value on F test of 42 school -- -- -- fixed effects Adj. [R.sup.2] (Pseudo-[R.sup.2] 0.326 0.387 0.398 in column ) N 564 564 362 Note: t statistics based on robust standard errors in columns (5) and (6). Also included in all equations here and in Table 7 are: rank in the alphabet, indicators for female, English education, and econometrician, and a quadratic in years since the first paper. Columns (5) and (6) also include the number of full professors. TABLE 7 Testing for Informational and Other Determinants of Salary Effects, 43 Public-University Economics Departments Interacted With: (1) (2) Ind. Var. Use Citations Broad College Total citations rank/100 0.0406 0.0469 (4.71) (4.63) Total citations rank/100x interaction 0.0110 -0.0083 (0.59) (0.54) Citations to most-cited paper rank -0.0114 -0.0173 /100 (1.64) (2.14) Citations to most-cited paper rank/ -0.0059 0.0098 100x interaction (0.40) (0.80) No. of entries rank/100 0.0087 0.0100 (2.10) (2.13) No. of entries rank/100x interaction -0.0034 -0.0053 (0.39) (0.74) Main effect -- -- F statistic on interactions 0.15 0.43 Df (3,509) (3,509) Adj. [R.sub.2] 0.532 0.534 N 564 564 Interacted With: (3) (4) (5) Ind. Var. Moved NFULLS Ever Admin Total citations rank/100 0.0389 0.0623 0.0460 (4.07) (3.48) (5.40) Total citations rank/100x interaction -0.0035 -0.0011 -0.0284 (0.12) (1.22) (1.64) Citations to most-cited paper rank -0.0129 -0.0281 -0.0129 /100 (1.63) (1.89) (1.88) Citations to most-cited paper rank/ 0.0194 0.0001 0.0067 100x interaction (0.83) (1.13) (0.47) No. of entries rank/100 0.0118 0.0013 0.0074 (2.59) (0.14) (1.84) No. of entries rank/100x interaction -0.0026 0.0004 0.0101 (0.16) (0.85) (1.18) Main effect 0.184 -- -0.0174 (2.37) (2.36) F statistic on interactions 0.88 0.50 3.26 Df (3,306) (3,509) (3,506) Adj. [R.sub.2] 0.549 0.533 0.549 N 362 564 562 Note: Columns (1), (2), and (4) also include school fixed effects.
|Printer friendly Cite/link Email Feedback|
|Author:||Hamermesh, Daniel S.; Pfann, Gerard A.|
|Date:||Jan 1, 2012|
|Previous Article:||2010 Editor's report.|
|Next Article:||Should we get married? The effect of parents' marriage on out-of-wedlock children.|