A study of publication trajectories of the Brazilian Computer Science community

Abstract

The average faculty productivity have been described as a rapid rise-short peak-gradual decline pattern. Way et al. (2017) have studied this pattern for faculty careers in Computer Science in North America using a piecewise linear model. In this paper, we use a similar methodology and study trajectories (N = 20655) of the Brazilian Computer Science community. First, we have evaluated how the median publication count of researchers is related to institution prestige and public vs. private administration. Second, we have studied how the annual publication rates have increased and its variation according to prestige ranks of institutions. Third, we have found the average trajectory can indeed be described as the canonical rapid increase and slower decrease in productivity. For individual trajectories of senior researchers we have observed only 4.5% of trajectories are well explained by the conventional narrative of rapid rise and gradual decline model. We also have found polynomial models of degrees 1 to 3 explain almost 63.1% of trajectories. The rest of trajectories are considered unstable and not well explained by neither of approaches.

Full text

Anais da Academia Brasileira de Ciências (2019) 91(3): e20180559
(Annals of the Brazilian Academy of Sciences)
Printed version ISSN 0001-3765 / Online version ISSN 1678-2690
http://dx.doi.org/10.1590/0001-3765201920180559
www.scielo.br/aabc | www.fb.com/aabcjournal
An Acad Bras Cienc (2019) 91(3)
SOCIAL SCIENCES
A study of publication trajectories of the Brazilian Computer Science community
MARCELO K. ALBERTINI, ANDRÉ R. BACKES and ADRIANO L. DE SÁ
Faculdade de Computação, Universidade Federal de Uberlândia, Campus Santa Mônica, Av.
João Naves de Ávila, 2121, Santa Mônica, 38400-902 Uberlândia, MG, Brazil
Manuscript received on June 6, 2018; accepted for publication on September 17, 2018
How to cite: ALBERTINI MK, BACKES AR AND SÁ AL. 2019. A study of publication trajectories of the Brazi-
lian Computer Science community. An Acad Bras Cienc 91: e20180559. DOI 10.1590/0001-3765201920180559.
Abstract: The average faculty productivity have been described as a rapid rise-short peak-gradual decline
pattern. Way et al. (2017) have studied this pattern for faculty careers in Computer Science in North
America using a piecewise linear model. In this paper, we use a similar methodology and study trajectories
(N = 20655) of the Brazilian Computer Science community. First, we have evaluated how the median
publication count of researchers is related to institution prestige and public vs. private administration.
Second, we have studied how the annual publication rates have increased and its variation according
to prestige ranks of institutions. Third, we have found the average trajectory can indeed be described
as the canonical rapid increase and slower decrease in productivity. For individual trajectories of senior
researchers we have observed only 4.5% of trajectories are well explained by the conventional narrative
of rapid rise and gradual decline model. We also have found polynomial models of degrees 1 to 3 explain
almost 63.1% of trajectories. The rest of trajectories are considered unstable and not well explained by
neither of approaches.
Key words: Plataforma Lattes, faculty productivity, publication count, Brazilian Computer Science
community.
Correspondence to: Marcelo Keese Albertini
E-mail: albertini@ufu.br
ORCid: https://orcid.org/0000-0002-1846-946X
INTRODUCTION
There are three major foci on analysis of curriculum
vitaes (Cañibano and Bozeman 2009) career
trajectories, mobility and study of collective work.
Bayer and Dutton (1977) have studied productivity
trajectories as a function in time. Let y represent
productivity and x be time. A researcher can keep a
steady growing productivity as a function
~ yx
; a
function of declining rate of productivity
~ y log x
; a spurt function
32
~ yx x x−+
for a trajectory
with a burst of productivity; or an obsolescence
function with a burning-out phase of rapid decreasing
2
~y x xa− ++
.
Factors cited for productivity changes include
seniority eects, cognitive decline with age, and
formation of research teams (Levin and Stephan
1991, Fortunato et al. 2018). Recently Way et al.
(2017) argued individual trajectories for Computer
Science market in North America are better
explained by a piecewise linear model composed
of 2 linear functions than by a rapid-increase, short-
peak and gradual decrease, a conventional narrative
described by an obsolescence function.
In comparison to North America, the Brazilian
Computer Science community is underdeveloped.
It has been less than 40 years since the rst PhDs

MARCELO K. ALBERTINI et al. PUBLICATION TRAJECTORIES OF THE BRAZILIAN CS COMMUNITY
An Acad Bras Cienc (2019) 91(3) e20180559 2 | 12
working in this eld were granted their degrees
(see Figure 2 below). Additionally, few studies on
the career trajectory of Brazilian researchers can be
found. Scarpelli et al. (2008) studied the prole of
the Brazilian researchers in Dentistry which were
granted scholarships as a recognition for their high
productivity from the Brazilian National Research
Council (CNPq). This study indicated that 90.3%
of the granted researchers developed their activities
in public institutions and only 9.7% in private
institutions. Most of them, 75.0%, worked in the
State of São Paulo. Researchers awarded with
category 1 grants, which are of the highest prestige,
were found to account for 53.5% of production,
while researchers in category 2 accounted for 45.1%
of production. Santos et al. (2012) also applied
similar methodology to learn the distribution and
imbalance of grants to researchers in a subeld of
studies about oral pathologies. They have learned
only three institutions (Universidade de São
Paulo, Universidade Federal de Minas Gerais and
Universidade de Campinas) concentrated almost
half of higher productivity researchers. Picinin
et al. (2016) analysed productivity grants in the
context of production engineering field. They
studied individual trajectory profiles of granted
researchers and concluded that there is a logical
coherence regarding distribution of grants.
Mendes et al. (2010) conducted studies to
evaluate the productivity of scholars in the eld of
Medicine. The studies analyzed trajectories of 383
CNPq productivity grant researchers from 2005
to 2007. Such studies pointed out that among all
researchers, 65.9% were male, 72.8% worked on 2
states of the 27 federative units in Brazil, and 97.1%
are employed in universities. The concentration of
researchers in a few Brazilian States, namely São
Paulo, Rio de Janeiro and Minas Gerais, is a well-
known fact (Mugnaini et al. 2004).
In 2000, Guimarães et al. (2010) found more
than 70% of PhDs working in Brazil obtained their
degrees in Brazilian institutions. In three elds,
however, most of PhDs working in Brazil had
their degrees abroad: Theology (76.4% abroad),
Aviation and Aerospace Engineering (70.5%) and
Computer Science (54.2% abroad). International
cooperation is often linked to the evolution of
publication activity (Glänzel et al. 2006).
In Figure 1, our data shows the trend of PhDs
graduated abroad in Computer Science reduced in
2011 to 35.1% among faculties aliated to PhD
programmes. This change is due to new policies
for concession of PhD scholarships abroad aiming
to reduce costs and to avoid problems related to
migration of researchers to other countries (Ramos
and Velho 2011).
Data from 2010 shows the evolution of
postgraduate programs has played a key role to
the growth of scientic knowledge in Brazil (de
Almeida and Guimarães 2013). Mena-Chalco et al.
(2012) evaluated the prole of the bibliographical
productions of Brazilian programmes in Computer
Science from 2004 to 2009. They highlighted
the Brazilian computing area has a preference
towards publishing in conference proceedings
(approximately 71.5% being full papers, short
papers and abstracts), followed by journals
(approximately 15.9%) and 12% of books, book
chapters and others. Silva et al. (2017) also studied
production and collaboration patterns of Brazilian
CS programs. Another characteristics of Computer
Science community is its high inequality of gender,
with the prevalence of male researchers (Arruda
et al. 2009). Interestingly, female researchers are
more likely to have a higher-rank of productivity
grant.
In this study, we characterize the trajectory
of Computer Science researchers in Brazil,
similarly to the work of Way et al. (2017) on
the North American case. Although our data
structurally diers from their work, we were able
to draw comparisons between Brazilian and North
American research systems. In our study we have
taken into consideration the following important

MARCELO K. ALBERTINI et al. PUBLICATION TRAJECTORIES OF THE BRAZILIAN CS COMMUNITY
An Acad Bras Cienc (2019) 91(3) e20180559 3 | 12
Brazilian characteristics: academic tenure
probationary period of 3 years and tenure-track
faculties are rarely denied tenure1; research usually
1 This is distinct from US and Canada tenure in which the
duration of tenure track and rate of approval can vary. Some
reports states an average of 6 years to tenure be approved
(Batterbury 2008). These distinction of tenure-track may
is performed in public institutions (Mugnaini et al.
2004, Mendes et al. 2010, Santos et al. 2012); only
5 among the 50 institutions with highest prestige
are private ((Ranking Web of Universities 2017);
and most of PhDs work in public institutions (86%
aect patterns of faculty productivity.
Figure 2 - Growth of number of doctorate degrees since the end of decade of 1980. Dots
indicate number of doctorate degrees granted each year among those who declared to work
in Computer Science. Curve labeled with “CS” indicates number of Computer Science
degrees and “all” is unrestricted among those who work in the Computer Science eld.
Figure 1 - Origin of PhDs for faculties working in Computer Science PhD programmes according
to data from Brazilian Ministry of Education.

MARCELO K. ALBERTINI et al. PUBLICATION TRAJECTORIES OF THE BRAZILIAN CS COMMUNITY
An Acad Bras Cienc (2019) 91(3) e20180559 4 | 12
according to data collected for this study — see
Table II).
Another distinction between Brazilian and
North American systems is that in order to advise a
PhD student, a Brazilian tenured faculty often need
to be accepted into, rst, a Master Science and, later,
a PhD programme. Each programme can establish
its own acceptance guidelines. However, in a few
places, this acceptance is automatic as soon as the
faculty is hired and in others a researcher may need
to wait at least 3 years in order to have succesfully
advised undergraduate and master students.
The remaining sections of the paper are
organized as follows: Section “Data” describes how
we collected, structured and analysed our data on
productivity and prestige of institutions. In Section
“Publication Rates and Prestige”, we evaluate
overall productivity growth over time and study
how productivity of early career of researchers and
median publication count are related to prestige
of institutions. Section “Analysis of Average
and Individual Trajectories” presents a study of
models of average and individual productivity of
trajectories of at least 30 years of work. The nal
section draws conclusions about the current state of
the Brazilian Computer Science community.
DATA
The Brazilian Ministry of Science, Technology and
Inovation developed and maintains the Plataforma
Lattes2, a website used to store and publish academic
resumes. Researchers must host their resumes
in Plataforma Lattes to submit grant requests to
Brazilian agencies. The study of the impact of
Brazilian researchers often use Plataforma Lattes to
compute productivity metrics, for example, total of
production, total of citations, and h-index (Wainer
and Vieira 2013, Leite et al. 2011).
2 Plataforma Lattes: http://lattes.cnpq.br.
TABLE I
Summary of statistics about collected Curriculum Vitaes.
Prole Counting
CV’s (any type) stated to work in CS 20655
CV’s (CS-only) aliated to a top-200
institution
7992
PhDs (any type) working in CS 6605
PhDs (any type) aliated to a top-200
institution
4370
PhDs (CS-only) working in CS 2849
PhDs (CS-only) aliated to a top-200
institution working in CS
2026
TABLE II
Distribution of doctorates among type of institutions and
prestige ranks. Percentage of PhDs refers to number of
CV per institution, per rank.
Rank CV’s Institutions % PhDs
Private Institutions
1st-50th 410 5 57.5
51th-100th 594 20 46.8
101th-200th 320 59 29.3
Public Institutions
1st-50th 5485 45 58.8
51th-100th 823 30 50.3
101th-200th 360 42 33.8
We also accessed data from Sucupira System
organized by the Ministry of Education, from
which we identied who are the current faculties in
Computer Science doctorate programmes3.
Way et al. (2017) used data for the 2011-
2012 academic year from faculties given three
restrictions: they work in one of the 205 North
American PhD programmes; they obtained their
PhDs in North America; and they got their rst
assistant professorship in one of these institutions.
Therefore, they selected 2583 faculties in their
study.
In Brazil, for the 2018 academic year, there
were 31 PhD programmes in Computer Science,
3 Available at https://sucupira.capes.gov.br . Date of Access:
March, 13th 2018.

MARCELO K. ALBERTINI et al. PUBLICATION TRAJECTORIES OF THE BRAZILIAN CS COMMUNITY
An Acad Bras Cienc (2019) 91(3) e20180559 5 | 12
in which 1072 faculties were aliated as regular
members. However, we have identied only 181
among all 1072 faculties adhere to those same three
conditions chosen by Way et al. (2017). Still, only
90 of them have a career of at least 10 years, which
is a condition used by Way et al. (2017) to estimate
their piecewise model parameters.
Another peculiarity of the work by Way et
al. (2017) is their usage of a hiring-based prestige
previously introduced by Clauset et al. (2015). The
hiring-based prestige is related to the proportion of
doctorates from a institution to be hired as professor
in a high-prestige institution. For the Brazilian case,
considering faculties hired by rst time up to 2012,
we found only 12 institutions whose PhDs were
awarded and hired by the 31 PhD programmes
available in 2018.
Therefore, due to the low number of faculties
and PhD programmes, instead of reproducing
exactly the same analysis of Way et al. (2017) paper,
we extended our data to any person with resume
stored in Plataforma Lattes who declared Computer
Science to be a subject of interest (N=20655).
Consequently, we could not replicate all details
of their analysis. One important dierence is the
denition of length of career, which we dened
as the number of years from the first academic
publication, whereas Way et al. (2017) dened as
number of years from the rst hiring as professor
in a university.
The methodology of data collection
followed a breadth-first traversing of the graph
of collaborators declared by researchers in their
curriculum vitae and used the resumes of faculties
of some computer science departments in Brazil
as starting vertices. Researchers aliated to the
following institutions were used: Universidade de
São Paulo, Universidade Federal de Juiz de Fora,
Universidade Federal de Lavras, Universidade
Federal de Minas Gerais, Universidade Federal
de São João del-Rei, Universidade Federal de
Uberlândia, Universidade Federal de Viçosa,
Universidade de Campinas, Universidade Federal
de Alfenas, Universidade Federal de Itajubá. While
this list is not comprehensive, we found a graph
component with 564495 vertices from which we
selected N=20655 resumes who declared to work
in the eld of “Computer Science” (CS) which also
updated CV information from 2012 onwards. The
algorithm of breadth-rst traversing stopped when
no new collaborators were found. This collecting
process took place from mid-2013 up to early
2014, when Plataforma Lattes introduced a captcha
system to avoid automatic data retrieval. We used
GNU/Linux scripting tools to download (wget)
curriculum vitaes’ pages and customized code to
parse and look for links of collaborators’ pages.
In Plataforma Lattes, each person declares
a list of fields in which his activities are better
represented and links the resumes of the main
collaborators. Note there is no data about which
eld is their main one. Table I presents statistics
regarding the proles of resumes considered in this
paper. In this table, PhD students are also included.
In the collected data there are: full name,
affiliation details, fields of activity, and
bibliographic data on papers, books, chapters and
other publication items. In this paper, except when
otherwise stated, we only counted as publications
the following items: papers in journals, books,
chapters and full papers in proceedings. All
publications date from 1956 to 2012.
Note that, dierently from data used by Way
et al. (2017), these 20655 researchers are not
only faculties as they can also be technicians and
students. Therefore, we have analyzed publication
records using the year of rst publication which
can dier from the year of rst hire.
There is also data on academic degrees and
year of conclusion of PhD with its title (e.g.
Doctor of Science, Dr. rer. nat., PhD in Computer
Science, and PhD Electric Engineering). From
those who declared to work in Computer Science,
we have found 6605 doctorates. From those, 2849

MARCELO K. ALBERTINI et al. PUBLICATION TRAJECTORIES OF THE BRAZILIAN CS COMMUNITY
An Acad Bras Cienc (2019) 91(3) e20180559 6 | 12
have earned a doctorate degree title with words
containing prexes of “comput” or “inform” stems.
Throughout this paper, we used this method to
identify Computer Science (CS) degrees.
The curve of the years of conclusion of PhDs
is shown in Figure 2. In this figure, the curve
labeled with “all” includes all types of degrees and
the curve “CS” refers the counting of Computer
Science PhDs. These curves show the CS research
in Brazil has begun its growth during the decade
of 1980.
Due to the lack of data for computing a prestige
ranking based on hiring following methodology of
Clauset et al. (2015), instead we used the list of
200 best-ranked academic institutions according
to Webometrics Ranking of World Universities4.
The Webometrics ranking uses indicators based
on link analysis of institutional web pages and
bibliographic citation analysis. Combining data
from Plataforma Lattes and the Webometrics
Ranking, we have identified 7992 researchers
working in Computer Science in the 200 best-
ranked institutions in Brazil.
There are dierences in productivity due to
prestige. We have stratied institutions into three
rank levels: level 1 — from 1st to 50th, level 2 —
rank 51st to 100th, and third level — 101st to 200th.
Table II shows distribution of doctorates according
to the prestige groups and type of institutions.
This distribution is important to characterize the
role of public and private institutions in research.
The number of doctorates decreases according to
prestige level. Also, the proportion of doctorates in
public institutions is higher than private institutions.
In Figure 3, the average publication counts per-
person in institutions in each level are shown.
Among all Brazilian institutions considered,
only one institution is among the world top 200:
Universidade de São Paulo (world rank 63). The
4 Ranking available at http://www.webometrics.info/en/
Latin_America/Brazil.
number of doctorates aliated to Universidade de
São Paulo is 685.
As argued by Way et al. (2017), it would be
misleading to compare a 1960 publication with
a 2012 publication as the rate of production
has increased with time. This phenomena is also
noticed in Figure 4. In the following analysis,
publication counts are reported as taking 2012
as year of reference using the same equations as
economists use to compute the value of the dollar
of 1960 in 2012. For each year Y, its ination r(Y) is
calculated using
( )
2012c
(2012), the average publication
count per person in 2012, and the corresponding
average counting
( )
cY
of year Y according to
Equation 1. Figure 3 shows average publication
rates per-person not adjusted to ination, according
to prestige ranks. As mentioned by (Arruda et al.
2009, Cabanac et al. 2015, Cavero et al. 2014),
counting the total of publications can suer from
overcounting as a publication is counted as many
times as the number of authors. This aects Figures
3 and 4. However, this overcounting effect is
attenuated in our analysis of individual trajectories
by the ination correction.
( ) ( ) ( )
( ) ( )
2012
1 1
c cY
rY cY
−
= +
In addition, in 2001, Guimarães et al. (2010)
found out just over half of Brazilian Computer
Science researchers trained abroad. By comparison,
the majority of North American faculty is trained in
North American and originate from a small number
of prestigious universities, while in Brazil 34.9,
65% of faculties in PhD programmes is graduated
abroad.
Although the number of PhDs concluded
abroad has reduced, we found there are still
differences between these two classes. For
example, considering the CNPq productivity
grant distribution among researchers affiliated
to graduate programmes, the proportion of those

MARCELO K. ALBERTINI et al. PUBLICATION TRAJECTORIES OF THE BRAZILIAN CS COMMUNITY
An Acad Bras Cienc (2019) 91(3) e20180559 7 | 12
graduated abroad with a grant is 45%, while for
those graduated in Brazil is 35.4%.
Therefore, researchers graduated abroad are
more likely to obtain a higher category productivity
grant than those graduated in Brazil.
PUBLICATION RATES AND PRESTIGE
The Brazilian Computer Science research
community has begun to grow during the decade
of 1980, when the number of PhDs has increased
(Figure 2). This growth, seen in other elds (de Meis
et al. 2007), also aects the number of publications
each year, shown in Figure 4. From 1980 to 2000,
the publication count approximately doubles every
4.5 years. This rate of growth changed around
2004, when publication counts grew only linearly.
This is highlighted by the fact that from 2001 to
2004 (4 years), the publication count increased
58% (9677 more publications), while from 2004 to
2012 (9 years), it increased only 31% (8183 more
publications).
Previous studies have found that researchers
from more prestigious institutions tend to be more
productive. Way et al. (2017) found the faculty
in the 50-highest prestige institutions in North
Figure 3 - Average publication rate per-researcher according to the prestige rank of their institutions.
This gure counts publications in CVs of researchers with PhD in Computer Science which declared
aliation to a top-200 institution. This rate increases around 1995 due to the growth of the number
of PhDs.
Figure 4 - Total number of publications by those who declared to work in Computer Science. This
counting considers books, book chapters, articles in journal and publications (articles, extended
abstracts, and abstracts) in conference proceedings.

MARCELO K. ALBERTINI et al. PUBLICATION TRAJECTORIES OF THE BRAZILIAN CS COMMUNITY
An Acad Bras Cienc (2019) 91(3) e20180559 8 | 12
America have, in average, 30% higher production
rate per-person than those in positions 101 to 200.
In our case, the institutions in rst level of prestige
ranks have production 38.36% higher than those
in the second level and 73.7% higher than the
institutions in positions 101 to 200.
Figure 5 shows median publication counts
of doctorates correlated with institution prestige.
Our data corroborates this finding as indicated
by positive slopes of 0.19 for public and 0.06 for
private institutions.
The type of institution is relevant to productivity
analysis as among the 50 most prestigious Brazilian
institutions only 5 are private. This contrasts with
the scenario in North America, where the number
of institutions in the highest prestige level is about
50% for each of the types. Additionally, among the
institutions with the 50-highest median count only
11 are private.
There are 35 institutions in the highest level
among the 50 institutions with the highest median
count, while 13 institutions with highest median
are in the 51-100 prestige level.
Way et al. (2017) have identied for North
American institutions that publication counts over
the early career correlate with prestige but not with
the type of institution. In Brazil, we have identied
a weak correlation of publication counts with
prestige and also with the type of institution.
However, this observation changes when
considering productivity over entire careers of
PhD holders as seen in Figure 5. In this case, the
regressions show a stronger correlation of median
publication counts with prestige for public institutions
Figure 5 - Median publication count of doctorates correlated with institution prestige. Points
indicate the median number of publications adjusted for ination. Lines are robust linear regressions
(Venables and Ripley 2002) weighted by the number of researchers in each institution. This kind
of linear regressions provided by package MASS in R project is used to avoid outliers and follow
tendencies according to the number of researchers in each institution.

MARCELO K. ALBERTINI et al. PUBLICATION TRAJECTORIES OF THE BRAZILIAN CS COMMUNITY
An Acad Bras Cienc (2019) 91(3) e20180559 9 | 12
(slope equals to 0.19) than private ones (slope equals
to 0.06). Among the researchers who do not hold a
PhD degree, the median publication count does not
correlate with prestige nor type of institution.
ANALYSIS OF AVERAGE AND
INDIVIDUAL TRAJECTORIES
Figure 6 shows the trajectories of careers in top-50
institutions follow the conventional peak-decline
pattern. Although less individual trajectories are
available, this type of pattern seems also to be
present for senior researchers with longer careers.
On the other hand, in the analysis of
individual trajectories, we nd a higher variability
of productivity patterns. Figure 7 shows a
concentration of peaks of productivity around the
10th year of career among the 1000 most productive
researchers. This is similar to the tendency shown
by Way et al. (2017) for the North American case,
although the average of most productive year in our
case is shifted by about 5 years. We attribute this
dierence to our denition of the length of career.
Way et al. (2017) characterized the productivity
pattern within an individual career as a piecewise
model, which is presented in Equation 2. The
piecewise model uses two linear functions with
slope coecients 1
m
and 2
m
and a point of change
*
t
used by an indicator function
*
()It t>
, and
1
b
is the initial productivity.
( )
( ) ( )
( )
**
11 2 . 2f t b mt m t t I t t=++ − >
In order to understand the publication
trajectories of senior researchers, we have also
included polynomials of higher degrees (linear,
quadratic, cubic, up to 6th degree) in our pool of
models.
Figure 8 presents a visualization of trajectories
using 1
m
and
2
m
coecients from Equation 2.
In this gure, trajectories are marked by symbols
(circle, triangle, square, plus sign, and crossed
square) which identify the best-matching model
according to the Bayesian Information Criterion
(BIC) (Schwarz 1978). This criterion is useful to
penalize overly-complex models.
Each trajectory is tted using all models, which
are the piecewise-linear model and the polynomials
where degree 1 up to 6, and compared with the
BIC values. If the best BIC value of a model for a
trajectory is a polynomial with degree 4 or higher,
it is marked as unstable.
Figure 6 - Average trajectories of researchers in top 50 institutions with career length of 10 to 25
years follow a peak-decline shape.

MARCELO K. ALBERTINI et al. PUBLICATION TRAJECTORIES OF THE BRAZILIAN CS COMMUNITY
An Acad Bras Cienc (2019) 91(3) e20180559 10 | 12
Figure 7 - Heat map of the peak-year for the 1000-most productive researchers in the Brazilian
Computer Science community.
Figure 8 - Visualization of trajectories using coecients of a piecewise linear function for career
trajectories of senior researchers (288 trajectories). Each trajectory is marked by the best-tting
model according to BIC value. The piecewise-linear function had the smallest BIC for 58 (20.1%)
trajectories. Only CN = 70 are within the octant corresponding to the conventional narrative, that
is
1
0 m>
, 2
0 m
<, and
12
mm>
. Among the trajectories in CN, only 4.5% (N=13) are best-t with
piece-wise model. The number of trajectories in the quadrants are Q1 = 15, Q2 = 77, Q3 = 33, and
Q4 = 163.

MARCELO K. ALBERTINI et al. PUBLICATION TRAJECTORIES OF THE BRAZILIAN CS COMMUNITY
An Acad Bras Cienc (2019) 91(3) e20180559 11 | 12
The quadrants Q1, Q2, Q3 and Q4 allow
checking how the trajectories are evolving according
to the piecewise model. The shadowed area, an octant
indicated by CN, implies a starting positive slope
followed by a decreasing slope which represents the
conventional narrative of the academic trajectories:
a rapid-rise and gradual-decline pattern.
In the study of the trajectories, we have
found the linear model best ts the largest number
of trajectories (91 trajectories). The piecewise
linear model is the second most common with
58 trajectories. However, the number of unstable
trajectories is signicant (65 trajectories).
In Way et al. (2017) work, they have identied
44.9% of trajectories as linear and 20.3% as following
the conventional narrative. In our data, only 31.5%
are linear and 4.5% are conventional narrative.
RESEARCH IMBALANCE
Individual trajectories can be aected by factors
linked to the years during which researchers started
their careers (Way et al. 2017). For example, Way
et al. (2017) observed that faculties among those
first hired during the 1970s, 50% of academic
publications of their 5 rst years were produced by
the 15% most productive.
In our data collected from the Brazilian
CS community, taking into account the ination
correction of publication counts, we have found
a weak correlation between production imbalance
with seniority. The Gini’s index for productivity
among researchers who started their careers before
1980s is G=0.44. For the 1980s, the imbalance
increased to 0.47, and, for the 1990s lowered to
0.46 and, for the 2000s, the imbalance increased to
G=0.48.
The imbalance of 2000s indicates
approximately 50% of the production among those
researchers is attributed to the 20% most productive.
Although these indexes over time represent a small
increase in imbalance, it may indicate a tendency
of concentration of publications among fewer
researchers. We believe further data is needed to
understand this tendency, which could be explained
by several factors, among those: competition among
senior and junior researchers for grants and students;
and overhead for young researchers who started their
careers in institutions in process of development.
CONCLUSIONS
In this paper, we evaluated characteristics of the
Brazilian Computer Science community. Using data
from Ministério da Educação and Ministério da Ciência
e Tecnologia, we found 20655 active researchers from
which about 38.7% are PhDs aliated to one of the
200 best-ranked Brazilian institution.
We found only 43.1% of PhDs awarded with
a title specic to Computer/Information Sciences.
The number of Brazilian CS PhDs has begun to
grow around 1995 which has also accelerated the
publication in the area. However, only in 2001
institutions has begun hiring more PhDs graduated
in Brazil than those graduated abroad. We noticed
one important distinction as PhDs with titles
awarded abroad (45%) are more likely to receive
CNPq productivity grants than PhDs titles awarded
in Brazil (35.4%).
Although Brazilian government has invested
in development of universities, CS academic
community in Brazil is still underdeveloped.
According to 2018 data, there are only 31 Brazilian
PhD programmes in comparison to 205 programmes
in North America. More specically, according to
data from the last ocial report regarding graduate
programmes, only 12 PhD programmes have
trained researchers which were hired among the
current 31 programmes.
Publication rate growth have followed PhD
conclusion rise in Brazil until 2005. We found that
the median production count is correlated with
prestige and also with the type of institution (private
or public). A counterintuitive fact discovered is
that researchers in their early career have higher

MARCELO K. ALBERTINI et al. PUBLICATION TRAJECTORIES OF THE BRAZILIAN CS COMMUNITY
An Acad Bras Cienc (2019) 91(3) e20180559 12 | 12
production median in private institution than in
public ones.
Similarly to Way et al. (2017) study, we
found individual researcher trajectories often do
not follow rapid-rise, short-peak gradual decline
pattern. Less than 5% of careers can be described
by the conventional narrative.
AUTHOR CONTRIBUTIONS
MKA and ALS collected the data and created the
gures. ARB and MKA performed data analysis
and wrote the paper. All authors contributed toward
interpreting the results, revising, and improving the
paper.
REFERENCES
ARRUDA D, BEZERRA F, NERIS VA, ROCHA DE TORO
P AND WAINERA J. 2009. Brazilian computer science
research: Gender and regional distributions. Scientometrics
79(3): 651-665.
BATTERBURY S. 2008. Tenure or permanent contracts in
North American higher education? A critical assessment.
Policy Futures in Education 6(3): 286-297.
BAYER AE AND DUTTON JE. 1977. Career age and research-
professional activities of academic scientists: Tests of
alternative nonlinear models and some implications for
higher education faculty policies. J of High Educ 48(3):
259-282.
CABANAC G, HUBERT G AND MILARD B. 2015. Academic
careers in computer science: continuance and transience of
lifetime co-authorships. Scientometrics 102(1): 135-150.
CAÑIBANO C AND BOZEMAN B. 2009. Curriculum vitae
method in science policy and research evaluation: the state-
of-the-art. Res Evaluat 18(2): 86-94.
CAVERO JM, VELA B AND CÁCERES P. 2014. Computer
science research: more production, less productivity.
Scientometrics 98(3): 2103-2111.
CLAUSET A, ARBESMAN S AND LARREMORE DB.
2015. Systematic inequality and hierarchy in faculty hiring
networks. Sci Adv 1(1).
DE ALMEIDA ECE AND GUIMARÃES JA. 2013. Brazil’s
growing production of scientific articles-how are we
doing with review articles and other qualitative indicators?
Scientometrics 97(2): 287-315.
DE MEIS L, ARRUDA AP AND GUIMARÃES J. 2007. The
impact of science in Brazil. IUBMB Life 59(4-5): 227-234.
FORTUNATO S, BERGSTROM CT, BÖRNER K, EVANS
JA, HELBING D, MILOJEVIC S AND BARABÁSI AL.
2018. Science of science. Science 359(6379).
GLÄNZEL W, LETA J AND THIJS B. 2006. Science in Brazil.
part 1: A macro-level comparative study. Scientometrics
67(1): 67-86.
GUIMARÃES R, LOURENÇO R AND COSAC S. 2010. O
perfil dos doutores ativos em pesquisa no Brasil. Parcerias
Estratég 6(13): 122-150.
LEITE P, MUGNAINI R AND LETA J. 2011. A new indicator
for international visibility: exploring Brazilian scientific
community. Scientometrics 88(1): 311.
LEVIN SG AND STEPHAN PE. 1991. Research productivity
over the life cycle: Evidence for academic scientists. Am
Econ Rev 81(1): 114-132.
MENA-CHALCO JP, DIGIAMPIETRI LA AND OLIVEIRA
LB. 2012. Perfil de produção bibliográfica dos programas
brasileiros de pós-graduação em ciência da computação.
Encontro Brasileiro de Bibliometria e Cientometria 3.
MENDES PHC, MARTELLI DRB, SOUZA WPD, QUIRINO
FILHO S AND MARTELLI JÚNIOR H. 2010. Perfil dos
pesquisadores bolsistas de produtividade científica em
medicina no CNPq, Brasil. Rev Bras Educ Méd 34(4):
535-541.
MUGNAINI R, JANNUZZI PDM AND QUONIAM L.
2004. Indicadores bibliométricos da produção científica
brasileira: uma análise a partir da base Pascal. Ciênc Info
33(2): 123-131.
PICININ CT, PILATTI LA, KOVALESKI JL, GRAEML AR
AND PEDROSO B. 2016. Comparison of performance of
researchers recipients of cnpq productivity grants in the
field of Brazilian production engineering. Scientometrics
109(2): 855-870.
RAMOS MY AND VELHO L. 2011. Formação de doutores
no Brasil e no exterior: impactos na propensão a migrar.
Educ Soc 32: 933-951.
RANKING WEB OF UNIVERSITIES. 2017. Ranking web
of universities. https://www.webometrics.info/en/detalles/
usp.br
SANTOS MIP, LEITE BGL, PARANAÍBA LMR, OLIVEIRA
EA, VERÍSSIMO FM, OLIVEIRA RADD AND
MARTELLI-JÚNIOR H. 2012. Profile and scientific
production of Brazilian researchers in oral pathology. Revi
Odontol UNESP 41(6): 390-395.
SCARPELLI AC, SARDENBERG F, GOURSAND D, PAIVA
SM AND PORDEUS IA. 2008. Academic trajectories of
dental researchers receiving CNPq’s productivity grants.
Braz Dent J 19(3): 252-256.
SCHWARZ G. 1978. Estimating the dimension of a model.
Ann Stat 6(2): 461-464.
SILVA THP, LAENDER AHF, DAVIS CA, DA SILVA
APC AND MORO MM. 2017. A profile analysis of
the top Brazilian computer science graduate programs.
Scientometrics 113(1): 237-255.
VENABLES WN AND RIPLEY BD. 2002. Modern applied
statistics with S (Fourth ed.). New York: Springer.
WAINER J AND VIEIRA P. 2013. Correlations between
bibliometrics and peer evaluation for all disciplines: the
evaluation of Brazilian scientists. Scientometrics 96(2):
395-410.
WAY SF, MORGAN AC, CLAUSET A AND LARREMORE
DB. 2017. The misleading narrative of the canonical
faculty productivity trajectory. Proceedings of the National
Academy of Sciences of the United States of America 1-8.