such factors as publication counts, measuring output, and citation counts, measuring impact (and derived measures such as indexes, discussed next).

While numeric criteria trigger strong reactions,^c alternatives have problems too: peer review is strongly dependent on evaluators’ choice and availability (the most competent are often the busiest), can be biased, and does not scale up. The solution is in combining techniques, subject to human interpretation:

5. Numerical measurements such as publication-related counts must never be used as the sole evaluation instrument. They must be filtered through human interpretation, particularly to avoid errors, and complemented by peer review and assessment of outputs other than publications.

Measures should not address volume but impact. Publication counts only assess activity. Giving them any other value encourages “write-only” journals, speakers-only conferences, and Stakhanovist research profiles favoring quantity over quality.

6. Publication counts are not adequate indicators of research value. They measure productivity, but neither impact nor quality.

Citation counts assess impact. They rely on databases such as ISI, CiteSeer, ACM Digital Library, Google Scholar. They, too, have limitations:

˲ Focus. Publication quality is just one aspect of research quality, impact one aspect of publication quality, citations one aspect of impact.

˲ Identity. Misspellings and mangling of authors’ names lose citations. Names with special characters are particularly at risk. If your name is Kröten-fänger, do not expect your publications to be counted correctly.

˲ Distortions. Article introductions heavily cite surveys. The milestone article that introduced NP-completeness has far fewer citations than a later tutorial.

˲Misinterpretation. Citation may imply criticism rather than appreciation. Many program verification arti-

c D. Parnas, “Stop the Numbers Game,” Commun. ACM 50, 11 (Nov. 2007), 19–21; available at http://tinyurl.com/2z652a. Parnas mostly discusses counting publications, but deals briefly with citation counts.

An issue of concern to
computer scientists
is the tendency to
use publication
databases that do not
adequately cover cS.

cles cite a famous protocol paper—to show that their tools catch an equally famous error in the protocol.

˲ Time. Citation counts favor older contributions.

˲ Size. Citationcountsareabsolute;im-pact is relative to each community’s size.

˲ Networking. Authors form mutual citation societies.

˲ Bias. Some authors hope ( unethically) to maximize chances of acceptance by citing program committee members.

The last two examples illustrate the occasionally perverse effects of assessment techniques on research work itself.

The most serious problem is data quality; no process can be better than its data. Transparency is essential, as well as error-reporting mechanisms and prompt response (as with ACM and DBLP):

7. Any evaluation criterion, especially quantitative, must be based on clear, published criteria.

This remains wishful thinking for major databases. The methods by which Google Scholar and ISI select documents and citations are not published or subject to debate.

Publication patterns vary across disciplines, reinforcing the comment that we should not judge one by the rules of another:

8. Numerical indicators must not serve for comparisons across disciplines.

This rule also applies to the issue (not otherwise addressed here) of evaluating laboratories or departments rather than individuals.

 

cS coverage in major Databases An issue of concern to computer scientists is the tendency to use publica-

Calendar
of Events

april 13–16 computation and control, San francisco, ca, Sponsored: SiGbed, contact: Paulo tabuada, email: tabauda@ee.ucla.edu

april 15–18
the 8th international
conference on information
Processing in Sensor networks,
San francisco, ca,
Sponsored: SiGbed,
contact: rajesh Gupta,
email: rgupta@ucsd.edu

april 20–24
design, automation and test in
europe,
nice, france,
Sponsored: SiGda,
contact: benini luca,
email: lbenini@deis.unibo.it

april 20–24
the 18th international World
Wide Web conference,
madrid, Spain,
contact: Gonzalo leon,
email: gonzalo.leon@pucp.
edu.pe

april 26–30
international conference on the
foundations of digital Games,
Port canaveral, fl,
contact: emmet James
Whitehead, Jr.,
email: ejw@cs.ucsc.edu

may 1–2

Western canadian conference on computing education, burnaby, bc canada, contact: diana cukierman, email: diana@cs.sfu.ca

may 5–8
14th international conference
on animation, effects, Games
and digital media,
Stuttgard, Germany,
contact: thomas haegele,
email: thomas.haegele@
filmakademie.de

may 10–13
acm 2009 international
conference on Supporting
Group Work,
Sanibel island, fl,
Sponsored: SiGchi,
contact: erling carl havn,
email: havn@man.dtu.dk

APriL 2009 | voL. 52 | no. 4 | communicAtionS of the Acm

33