a validation study of bibliometric indicators: the comparative performance of cum laude doctorates...
TRANSCRIPT
Scientometrics, Vol. 17, Nos 5-6 (1989) 42 7-435
A VALIDATION STUDY OF BIBLIOMETRIC INDICATORS: THE COMPARATIVE PERFORMANCE OF CUM LAUDE DOCTORATES
IN CHEMISTRY
A. J. NEDERHOF, A. F. J: VAN RAAN
Science Studies Unit, LISBON-Institute I University of Leiden, Stationsplein 242, 2312 AR Leiden (The Netherlands)
(Received January 17, 1989)
The validity of bibliometric indicators as a monitor of the impact and usefulness of scientific research is examined by compaiing the scientific performance of cure laude and non-cum laude degree holders in chemistry (N=237), from five y~ears before their graduation to four years afterwards. Papers of cum laudes were cited more frequently than those of non-cum laudes from three years before graduation until one year after graduation..Two to three years after graduation, the short-term impact per paper was no longer significantly different for both groups. A similar pattern was found with regard to productivity. Little evidence was found in favor of the Ortega hypothesis and the Matthew effect. The results support the concurrent validity of bibliometric indicators with peer review indicators of quality of the research project.
Introduction
During the past decade, the assessment of scientific performance has been studied
ex~ensively in scientometrics, resulting in an enhanced appraisal of the strengths
and liabilities of both peer review and bibliometric methods. Recently, limitations
of peer review have become apparent (Harnad, 1982; lrvine and Martin, 1983;
Moed, Burger, Frankfort and Van Raan, 1985a; Nederhof, 1988), while biometric
studies claim only to monitor the impact and usefulness of scientific research, and not
the quality of scientific performance (e.g., Moed, Burger, Frankfort and Van Raan, 1985b).
Critics of bibliometric methods have attempted to show that citations are only,
or in the first place, rethoric instruments used, not to convince others by the strength
of arguments but to persuade th0m by nonlogical means (Cozzens, 1989; Gilbert, 197 7 ; MacRoberts and MacRoberts, 1986). Also, MacRoberts and MacRoberts (1986:159) make a case for the importance of 'ceremonial citations', references made
to individuals whose work has actually not been influential, but to whom the citer
Scientometrics 17 (1989) Elsevier, Amsterdam-Oxford-New York Akad~rniai Kiad6, Budapest
A. J. NEDERHOF, A. F. J. VAN RAAN: VALIDATION OF BIBLIOMETRIC INDICATORS
feels obliged. In the present study, we will investigate whether or not a significant number of references are made for other reasons than persuasion or ceremonial purposes. We studied citations to the work of graduate students. When citations are purely made to persuade an audience, or for ceremonial reasons, it would be expected that the
work of graduate students, who will not be recognized very well or respected very highly by the scientific audience, would be cited rather low uniformly, especially
after removal of the effects of the presence or absence of prestigeous or powerful
co-authors. In addition, it would be expected that the post-Ph.D, work of graduates awarded with a 'cum laude' ("cumlaudes") would be cited more after their award than before, both in comparison with their own work, as well as with that of graduates not receiving their Ph.D. with a cure laude ("non-cumlaudes").
A simlar hypothesis could be deduced from the Matthew effect (Merton, 1968): before graduation, differences between cumlandes and noncumlaudes-to-be would be small, as grounds are lacking for those outside the particular university to distribute merit (i.e,, citations) according to (indirectly) observable distinctions, such as a better position. However, post-Ph.D, publications by cumlaudes should receive increasingly more citations than noncumlaudes, because of the cumulative nature
of the Matthew effect. According to the Ortega hypothesis (Ortega y Gasser, 1932), progress in experi-
mental science is due in great part to the work of mediocre and less than mediocre
scientists. Provided that scientists as a group distribute their citations according to
the contribution scientists actually have made to scientific progress, the Ortega hypothesis implies that a relatively large share of citations obtained by graduates would be received by non-cumlaudes, who should also produce a sizable share of
the highly cited publications. A fourth hypothesis assumes that both peer review and bibliometric measures
are partially valid indicators of past scientific performance. Therefore, it is expected that publications of cumlaudes are cited more often than publications of non, cumlaudes. When, additionally, the assumption is made that cumlaudes outperform
non-cumlaudes because they are better researchers, than it would be expected that
cumlaudes continue to outperform non-cumlaudes after graduation, and would also continue to produce more highly cited publications than non-cumlaudes.
A fifth hypothesis is that the quality of the research project, and not so much the quality of the particular student, is the most important factor in both produc- tivity levels of researchers and impact levels of scientific publications (Nederhof and Van Raan, 1987: 334). Then, cumlaudes would outvie non-cumlaudes, and also be authors of more highly cited publications, as long as their publications concern their graduate work. However, when cumlaudes and noncumlaudes embark
428 Scientometrics 1771989)
A. J. NEDERHOF, A. F. J. vAN RAAN: VALIDATION OF BIBLIOMETRIC INDICATORS
upon new'research projects after graduation, initial differences may disappear. In
our study of physics graduates, we indeed found this pattern of results (Nederhof and Van Raan, 1987).
To test the various hypotheses, productivity and impact of cum-laudes and non- cumlaudes in chemistry were studied from five years before graduation until three
years after graduation.
Method
The files of the Leiden Indicators Project were used as data source, especially publication and citation data of the chemistry departments at the University of Leiden for a fourteen-year period (1970-1983). These data, used in previous
studies of research performance, are amply discussed in earlier work of our group
(Moed et al., 1985a, 1985b). In-house citations (references given by other publica- tions with a Leiden address) were not included in the citation counts. To remove
potential differences due to type of publication, only research articles were included
in the publication file (the articles coded as a blank by ISI). Book publications were excluded. Regarding publications with multiple authors, only the status (cum
laude or not) was taken into consideration of those authors who could fulfill all
conditions posed to inclusion. Thus, if, for instance, only papers published before graduation are sonsidered in a particular analysis, the status of co-authors who have already graduated is disregarded. This was done in order to focus as much as
possible on the performance of the graduates, independently from co-authorships. The doctorate files of the University of Leiden have been searched for the data
concerning the cum laude award. In five chemistry departments, 28 cumlaudes (11,8%) and 209 non-cumlaudes (99,2%) were found.
The impact of publications was computed according tct'citation windows',
consisting of citations obtained in the year of publication of the paper and the two subsequent years.
Results
To examine the comparative changes in productivity and impact of the graduates over time, Table 1 was constructed.
From five years before graduation to three years ,after graduation, the produc- tivity of graduates, the short-term impact of their publications, as well as the total number of citations received were computed for eight consecutive partially over lapping two-year blocks (see Figs 1, 2, and 3).
Scientometrics 17 {1989} 429
A. J. NEDERHOF, A. F. J. VAN RAAN: VALIDATION OF BIBLIOMETRIC INDICATORS
Table 1 Short-term impact and productivity of cumlaude and non-cumlaude graduates in chemistry
Cumlaudes Non-cumlaudes Publication
blocks C/a P/a C/P C/a P/a C/P
Yg-5 to Yg-4 0.076 0.615 0.125 1.574 0.327 4.818 Yg-4 to Yg-3 1.687 0.375 4.500 1.403 0.468 3,000 Yg-3 to Yg-2 5.647 0.706 8,000 /..743 0.625 2.789 Yg-2 to Yg-1 7.400 1.700 4.353 2.451 0.915 2.680 Yg-1 to Yg 8.478 2.391 3,545 3.583 1.389 2,580 Yg to Yg+l 9.178 2.178 4.213 4.071 1.565 2.602 Yg+l to Yg+2 6.885 1.538 4,475 2.780 0.995 2.795 Yg+2 to Yg+3 2.375 0.667 3.563 1.384 0.469 2.95T
C/a = average number of citations per author; P/a = average number of publications per author; C/P = average number of citations per publication; Yg = year of graduation.
A
o Cumlaudes .No cu . ~ 6 ' 0
o ~ , r I [ I I I E I -5,-4 -z4-3. -3,-2 -2,-1 -1,0 0,1 ~,2 2,3 <
2-yeors publication blocks with Yg=0
Fig. 1. Citations per author for cumlaudes and non-cumlaudes over time
The first two year-block covers the f if th and four th year before graduation,
while the eighth b lock covers the second and third year after graduation. With
regard to ci ta t ion counts , blocks were formed consisting o f two consecutive
publicat ion years, each having its own 0 - 2 year ci ta t ion window.
The product iv i ty ' o f cumlaudes and non-cumlaudes is rather similar f rom five to
two years before graduation. Two years before graduation, however , the prodttc-
t ivi ty o f cumlaudes more than doubles f rom 0.7 paper to 1.7 paper, while this
430 Scientometrics 17 (1989)
A. J. NEDERHOF, A. F. J. VAN RAAN: VALIDATION OF BIBLIOMETRIC INDICATORS
~ 2/4
a. 1.92
.u 1Az
.D
o961 "6
r~
c
o
.~ - , ,
~ 4 I t I L L l -5,-z, -4,-3 -3,-2 -2,-1 -I,0 0,1 1,2 2,3
2-years publication blocks with Yg--0
Fig. 2. Publications per author for cumlaudes and non-cumlaudes
cz 6z Q.
"6 3.2
i 1.6
o
A o Cumlaudes
/ - ' " - f - i : - ' i .... .... -5,-~ -~,-3 -3,-2 -2,-1 -1,0 0,1 1,2 2,3
2-ye~s publication blocks with Yg:O
Fig. 3. Short-term impact per paper for cumlaudes and rton-cumtaudes
increase is less than 50% for noncumlaudes compared to the previous two-year
block. Productivity reaches a peak at 2.4 papers for cumlaudes in the year preceding
graduation and the graduation year itself. The productivity of noncumlaudes also
increases, to 1.4 papers on the average, one full paper below the level of cumlaudes.
The gap begins to close, however, during the year of graduation and the year
thereafter, as prodcutivity of cumlaudes declines somewhat to 2.2 paper, while
that of noncumlaudes increases further to 1.6 paper. Thereafter, the registered
productivity of both groups declines steeply, partly no doubt because of finding
Scientometrics 17 (1989) 431
A. J. NEDERHOF, A. F. J. VAN RAAN: VALIDATION OF BIBLIOMETRIC INDICATORS
jobs outside Leiden University. In the second and third year after graduation, the "Leiden" productivity of cumlaudes is only 0.2 papers above that of non-cumlaudes.
The picture changes somewhat when the total short-term impact is examined of all papers produced during two consecutive years. From five to four and from
four to three years before graduation, cumlaudes and noncumlaudes are cited about
equally (due to small numbers, the relatively high score of non-cumlaudes in the first two-year block is not significantly different from the relatively low score of the cumlaudes). Three to two years before graduation, however, the number of citations to cumlaudes soar significantly from 1.7 to 5.7 citations, while non- cumlaudes only slightly improve from 1.4 to t.7 citations. The published research
papers of cumlandes continue to draw more citations until the two year period
covering the graduation year and the subsequent year, when they are cited 9.2 times. In the same period, the publications of non-cumlaudes are also more
frequently cited, but at a much slower pace, increasing from 1.7 to a peak of 4.1. citations. In the two subsequent blocks, covering 1 to 3 years after graduation, the number of citations to Leiden publications of the graduates decrease dramatically, although cumlaudes continue to receive somewhat more citations than non-cumlaudes.
The most important measure of impact is the average number of citations per
paper. Interestingly, the few non-cumlaudes who publish early (5 to 4 years before
graduation) are never cited more frequently during their subsequent career. Due to
small numbers, the seemingly huge difference in average short-term impact per paper between non-cumlaudes (4.8 citations) and cumlaudes (0.1 citations) is not
significantly different (p < 0.06). Although the impact per paper is higher for cumlaudes (4.5 citations) than for non-cumlaudes (3.0 citations) 4 to 3 years
before graduation, this difference is also not significant. However, 3 to 2 years before graduation, papers of currdaudes are cited 8 times
on average (versus only 2.8 times for non-cumlandes, p < 0.004), a level not to be surpassed again during the period surveyed in this study. After that, papers of cumlaudes average between 3.5 and 4.5 citations each, compared to a significantly lower 216 to 2.8 citations for papers of non-cumlaudes until the second year after
graduation. Only in the last two-year block, extending 2 to 3 years after graduation, the
average number of citations per paper of cumlaudes (3.6 times) and non-cumlaudes (3.0 times) are not significantly different. It is important to check wether this is due the structure of our data file, which contains only publications with a Leiden
address. After graduation, many leave the Leiden University to get a job elsewhere. To get a better view of what happens after graduation, we compared the output
of first-authored cited papers produced two to four years after graduation of both cumlaudes and a matched sample (as regards year of graduation and mentor) of
432 Scientometrics 17 (1989)
A. J. NEDERHOF, A. F. J. VAN RAAN: VALIDATION OF BIBLIOMETRIC INDICATORS
non-cumlaudes in chemistry. Thus, persons no longer active in (academic) research
were excluded from the samples, and only papers were included which were likely
to be produced after graduation. Included papers were published between 1974 and
1980. Data were collected concerning 24 cumlaudes and 25 non-cumlaudes.
The results were as follows. Cumlaudes produced on average 0.65 papers per
year, while noncumlaudes averaged only half that amount: 0.33 papers. Papers by
cumlaudes were cited 6 times on average during the first three year, versus only 3.7
times when non-cumlaudes were authors. Although this concerns only first-
authored papers, these results suggest strongly that cumlaudes who continue
publishing after graduation, clearly outproduce active non-cumlaudes, and that
cited papers of cumlaudes have a higher impact on average than cited papers of
non-cumlaudes.
Number of highly cited papers per author
Another important way of measuring the impact of work is to look at the
number and percentage of papers exceeding a particular level of citations. We
examined the short-term impact of papers produced as a result of graduate work
by focussing on the period from 3 years before graduation to 1 year after. The
results are presented in Table 2.
Table 2 Number of (highly) cited papers per author for cumlaudes
and non-cumlaudes from three years before graduation to one year after graduation
Citation level/> Cumlaudes Non-cumlaudes
0 3.94 (100%) 2.90 (100%) 1 3.12 ( 79.1%) 1.89 ( 65%) 5 1.29 (32.8%) 0.53 (18.2%)
10 0.41 (10.4%) 0.17 (6.0%) 15 0.29 (7.5%) 0.08 (2.9%) 20 0.12 (3.0%) 0.05 (1.7%) 30 0.06 (1.5%) 0.01 (0.2%)
As has been shown already, cumlaudes are moreproduc t ive than non-cumlaudes.
The greater productivi ty of cumlaudes has not diluted the impact of their work, as
the average impact of their papers is clearly above that of non-cumlaudes. Whereas
79% of the papers by cumlaudes are cited, only 65% of the papers by non-cumlau-
des receive at least one citation. While cumlaudes produce on the average during
433
A. J. NEDERHOF, A. F. J. VAN RAAN: VALIDATION OF BIBLIOMETRIC INDICATORS
the five year period 1.3. papers which are cited five times or more (nearly a third of their total production), non-cumlaudes only produce 0.5 papers (representing 18% of their total production). On all citation levels examined, cumlaudes have
a higher percentage of their papers included than non-cumlaudes.
Discussion
In the present study, a comparison is made between the results of a peer review process, namely the awarding of a cure laude degree to graduates in chemistry, and bibliometric indicators of productivity and impact. By monitoring the work of the graduates from five years before graduation to three years after graduation, it was
possible to compare results with various hypotheses. The results do not fit the pattern predicted by the Matthew effect: little differ-
ence in short-term impact before graduation and an increasing difference between cumlaudes and non-cumlaudes after graduation. The actual pattern, however, is nearly reversed. Large difference in impact and productivity favor cttmlaudes three to two years before graduation, differences which decrease in the following years, although remaining significant. As eumlaudes and non-cumlaudes share the same mentors, the observed pattern cannot be attributed to differences in distiction of the co-authors. An alternative explanation might be, that cumlaudes reap the benefits of very early career awards, as for instance a cure laude undergraduate degree. The nonsignificant differ- ences in impact between 5 and 3 years before graduation do not point in this direction, however. We are therefore lead to conclude that a Matthew effect is not visible in our data. This replic~ites earlier results in a study of physics graduates (Nederhofand Van Raan, 1987). If a Matthew effect still exists, it takes longer than the period studied
here. According to the Ortega hypothesis, a relatively large share of citations and
highly cited publications are expected to be obtained by the relatively 'mediocre' non-cumlaudes. Indeed, of the 1476 citations received by graduates from three years before to one year after graduation, 1134 go to non-cumlaudes, and only 299 to cum-laudes. However, only 18% of the publications of non-cumlaudes draw 64% of all citations. These figures indicate that even among non-cumlaudes, a few persons attract the major share of citations. Thus, the Ortega hypothesis is hardly
supported by the present data. Various sceptics have claimed that bibliometric measures based on citations are
generally invalid. The present data do not offer any support for this stance. Highly significant differences in impact and productivity were obtained between two groups distinguished on a measure of scientific quality based on peer review (the cum laude award). As with other results of validation studies of citation studies, these
434 Scientometrics 17 (1989)
A. J. NEDERHOF, A. F. J. VAN RAAN: VALIDATION OF BIBLIOMETRIC INDICATORS
can no t be explained by sceptics. It is hard to see what differential ceremonial or
persuasive value a c i ta t ion towards the work o f a graduate s tudent would have.
Even though a percentage o f the ci tat ions may have been given for per func tory or
negative reasons, these results suppor t the con ten t ion that when sufficiently large
numbers o f papers are examined, ci tat ions counts may provide a useful partial
indicator o f quali ty, and can be used frui tful ly to mon i to r scientific research.
References
1. COZZENS, S. E. What do citations count? The Rhetoric-first model, Scientometrics, 15 (1989) 000.
2. GILBERT, G. N. Referencing as persuasion, Social Studies o f Science, 7, (1977) 113-122. 3. HARNAD, S. (Ed.), Peer commentary on peer review, Cambridge, Cambridge University Press,
1982. 4. IRVINE, J., MARTIN, B. Assessing basic research. Some partial indicators of scieritific progress
in radio astronomy, Research Policy, 12, (1983) 61 -90 . 5. MACROBERTS, M. H. MACROBERTS, B. R., Quantitative measures of communication in
science: A study of university research performance, Social Studies o f Science, 16 (1986), 151-172.
6. MERTON, R. K., The Matthew effect in science, Science, 159 (1986) 56-63. 7. MOED, H. F., BURGER, W. J. M., FRANKFORT, J. G., VAN RAAN, A. F. J., A comparative
study of bibliometric past performance analysis and peer judgment, Scientimetrics, 8 (1985) 149-159.
8. MOED, H. F., BURGER, W. J. M., FRANKFORT, J. G. VAN RAAN, A. F. J. The use of bibliometric data for the measurement of university research performance, Research Policy, 14 (1985) 131-149.
9. NEDERHOF, A. J. The validity and reliability of evaluation of scholarly performance, In: A. F. J. VAN RAAN (Ed.), Handbook o f Quantitative Studies in Science and Technology, Amsterdam, North Holland, 1988, pp. 193-228.
10. NEDERHOF, A. J. VAN RAAN, A. F. J. Peer review and bibliometric indicators of scientific performance: A comparison of cure laude doctorates with ordinary doctorates in physics, Scientometrics, 11 (1987) 333-350.
11. ORTEGA Y GASSET, J., The revolution of the masses. New York, Norton, 1932.
Scientometries 17 (1989) 435