leonardo reyes gonzalez evaluation of mexican groups
TRANSCRIPT
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
1/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
1
Evaluation of Mexican Research Groups in Physics and Mathematics
using an Endogenous Approach1
Leonardo Reyes-Gonzalez2
Francisco Veloso3
Abstract
Throughout the last three decades, tightening budgets, increasingcompetition and better understanding of the outputs of science and
technology have stimulated the development of new types of research
evaluations. Although useful, existing methods typically use ad-hocdefinitions of the unit of analysis, lacking approaches capable of
incorporating the self-organizing mechanisms of the research endeavor
into the evaluation effort.
This paper addresses this limitation by developing an evaluation method
that takes into account these endogenous characteristics of the researcheffort. In particular, the collaboration patterns of researchers are used toidentify the frontiers of the focal research units and the backward citation
patterns are employed to establish relevant benchmark units for each focal
unit.
This method is then tested with the fields of Physics and Mathematics inMexico in five leading institutions for the period of 1997-1999. Finally,
we present a detailed peer benchmark.
1 This work was supported the Mexican Council for Science and Technology (CONACYT) CMUs
Berkman Faculty Development Fund and the National Science Foundation (NSF)2Department of Engineering & Public Policy Program in Strategy, Entrepreneurship and Technological
Change, Carnegie Mellon University, [email protected]@cmu.edu, The usual disclaimers apply.
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
2/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
2
1. Introduction and Motivation
Throughout the last decades, tightening budgets and an increasing competition betweenresearch projects, combined with a higher awareness of the outputs on science, have
stimulated the development of new approaches towards research evaluation (COSEPUP,
1999; Georghiou and Roessner, 2000, van Raan, 2000; Rip, 2000; Frederiksen, Hansson
and Wenneberg, 2003). For example, current assessments have evolved from the classicalpeer review to an informed peer review, in which research is evaluated with the aid of
quantitative benchmarks; in the UK, the Research Assessment Exercises are tying
funding to output (publications) and recognition (citations) (Leydesdorff, 2005). Despite
an important evolution, evaluations still have a critical limitation: the boundaries of theunit of analysis are typically rigid (by institutes/departments, institutions, disciplines,
regions, or countries), overlooking the unique and self-organizing characteristics of the
research endeavor (Guimera et al., 2005). This means, for example, that present
techniques have difficulty noting differences between low and top performing groups
within a focal unit, say a university or even a department within a university. Likewise, benchmarking performance of university departments in a given area based solely on
number of papers or citations fails to recognize that theoretical or experimental researchprofiles will necessarily imply different levels of publication and citation outputs. This
renders a comparison based on average levels of productivity or impact for an area of
limited value and potentially misleading. Furthermore, the performance of groups, orsubunits (as described in Glser, Spurling and Butler (2004)) cannot be independently
measured. Finally, current methods are particularly limited when assessing
interdisciplinary research groups (RGs), because it is difficult to ascribe these groups to a
particular field of knowledge and measure their performance within this field incomparison with equivalent groups.
To overcome these limitations, we developed and test a research evaluationmethod that recognizes the endogenous, or self-organizing characteristics of research
groups. Instead of an ad-hoc definition of the unit of analysis, the proposed method will
use patterns of collaboration and the specific body of knowledge that these collaborationsentail to identify the frontiers of the focal units, as well as other units that qualify as
relevant benchmarks. First, the boundaries of a RG are be identified based on co-
authorship and using the notion of Cohesive Groups, a method used in Network Analysisfor subgroup identification (Wasserman and Faust, 1994, p. 249-284). Second, we use
backward citations (found in the work published by each group) to establish its
knowledge footprint. These footprints will be used to evaluate the degree of structural
similarity between groups, i.e. the similarity between RGs is to be defined by how muchtheir work cites common papers/journals. Once the RG are characterized and their peersidentified, expect to measure and benchmark the performance/productivity of each RG.
The method will be demonstrated by rankings group of the top five institutions in Physics
and Mathematics in Mexico
This paper is divided into five sections. First, we briefly describe different types
of evaluations and their limitations. Second, we make reference to the theories that
support our method, Bibliometric and Network Analysis. Third, we explain the method.Fourth, we apply this method to analyze the fields of Physics and Mathematics in Mexico
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
3/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
3
in 1997-1999 for five leading institutions and rank absolutely and relatively this area of
knowledge in this country, within this same time period. Finally we present some policyimplications.
2. Research Evaluation
According to Papaconstantinou and Poltto (1997, p. 10), evaluation refers to a processthat seeks to determine as systematically and objectively as possible the relevance,
efficiency and effect of an activity. The evaluation can occur at three different stages:
after (ex-post), during, or before (ex-ante) the activity. These evaluation types will
produce information that can be used in the assessment of past policies, the monitoringof ongoing initiatives or the forward planning of innovation and technology policies
(Papaconstantinou and Poltto, ibid).
As stated previously, in the last 30 years evaluations (in general) have been on the
rise. This trend has been fueled by budget stringency, the need to better allocate scarce public resources and by a broader reassessment of the appropriate role of government
(Papaconstantinou and Poltto, 1997, p. 9). With respect to Science and Technology(S&T), new developments in these areas, increasing costs, higher awareness of its effects,
and the desire to use the knowledge and outcomes of these activities, combined with the
need to understand the consequences of S&T polices have spurred the demand for thesetypes of activities (Martin and Irvine, 1983; COSEPUP 1999; van Raan, 2000; Rip, 2000,
Frederiksen, Hansson and Wenneberg 2003).
2.1 Types Research of Evaluations
To accommodate the changes and to address the new realities of S&T, several types ofevaluations have been developed. Table 1 lists and describes the most common types ofevaluations.
Table 1 Current Methods for Research Evaluation
Methods Description
Bibliometric
analysis
Assumes that publications, citations, and patent counts signal the
work and productivity of a unit of analysis.
Economic rate of
return
Used to estimate the economic benefits (such as rate of return) of
research; gives a metric for the outcomes of research.Peer Review
Classic approach Traditional method for evaluating science in which scientists
continuously exercise self-evaluation and correction. Focuses on
individual scientific products.
Modified
approach
Natural development from the classical one, it incorporatesissues that are not strictly cognitive. Focuses on group learning.
Case studies Historical accounts of the social and intellectual developments
that led to key events in science or applications of science
illuminate the discovery process in greater depth than other
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
4/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
4
methods
Retrospective
analysis
Similar to case studies, but instead of focusing on one scientific or
technological innovation it focuses on multiple cases.Benchmarking Used to assess whether a particular unit of analysis is at the cutting
edge in terms of research, education or other measures.
Source: COSEPUP (1999, p 18-22); van Raan (2000); and Frederiksen, Hansson and
Wenneberg (2003).
Each method is substantially useful in its own way, but has significant drawbacks
(Table 2). For all established methods noted above, important additional limitations exist.
Typically, the boundaries of the unit of analysis are defined in an ad-hoc way,overlooking the endogenous characteristics of such units. Furthermore, these evaluationsoften assign broad cohort groups and use a working assumption that within a field all
units are homogenous in terms of the knowledge they use and their comparative
advantage (e.g. differences between low and top performing actors within a given unit).
The method proposed in this study aims to address these limitations.
Table 2 Pros and cons in current evaluation methods
Methods Pro Con
Bibliometric
analysis
Quantitative; useful on
aggregate basis to evaluatequality for some programs
and fields
At best, measures only quantity; not
useful across all programs & fields;comparisons across fields or countries
difficult; can be artificially influenced
Economic rate
of return
Quantitative; shows
economic benefits ofresearch
Measures only financial benefits, not
social benefits; time separating researchfrom economic benefit is often long; not
useful across all programs and fields
Peer Review Well-understood method and practices; provides
evaluation of quality of
research and sometimesother factors
Focuses primarily on research quality;other elements are secondary; evaluation
usually of research projects, not programs;
great variance across agencies; concernsregarding use of "old boy network";
results depend on involvement of high-
quality people in processCase studies Provides understanding of
effects of institutional,
organizational, and technical
factors influencing research
process, so process can beimproved; illustrates all
Happenstance cases not comparableacross programs; focus on cases that
might involve many programs or fields
making it difficult to assess federal-
program benefit
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
5/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
5
types of benefits of research
process
Retrospectiveanalysis
Useful for identifyinglinkages between federal
programs and innovations
over long intervals of
research investment
Not useful as a short-term evaluation toolbecause of long interval between research
and practical outcomes
Benchmarking Provides a tool forcomparison across programs
and countries
Focused on fields, not federal researchprograms
Source: COSEPUP (1999, p 18-22).
3. Theoretical Background
In this section we describe the bodies of knowledge that have been used in the
development of our method. First we use dynamic network analysis to establish thepatterns of collaboration within a field and delimit the boundaries of an RG. In particular,
we draw on the notion of Lambda Sets to determine the level of cohesiveness of thesegroups. Second, we use two techniques from bibliometric analysis to establish the
similarities between RGs (co-citation analysis), assess their performance (with respect to
their publication output, citation counts, citation impact and citation impact by groupsize) and rank these groups against their peers.
3.1 Network Analysis
According to Rogers, Bozeman and Chompalov (2001), networks are used in the study of
science and technology (S&T) as guiding metaphors and as techniques to measure
structural properties of the ensemble. They classify these types of studies by the level
of analysis that is given by the nature of the actors that will be placed at the nodes (i.e.nodes can be individuals, teams, departments or institutions), by the nature of the links
between nodes (interaction networks vs. position networks)4, and by the domain in which
the actors belong (intra-organizational vs. inter-organizational)5. In this work, we use
network theory to understand the patterns of collaboration in science and delimit the
boundaries of an RG. For such purpose, we focus our analysis on the ties that emergethrough co-authorship and define an RG based on its levels of cohesiveness and the
connectivity between co-authors.
4 In Interaction networks, the links represent actual information exchanges or other communication events
between actors; whereas in position networks, they represent relationships established by the relative
positions of the actors in the system (Rogers, Bozeman and Chompalov, 2001)5Intra-organizational studies only consider links between actors inside the boundaries of a single
organization; in contrast, inter-organizational studies consider links between actors across organizationalboundaries (Rogers, Bozeman and Chompalov, 2001)
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
6/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
6
Cohesive subgroups
According to Wasserman and Faust (1994, p. 249), a cohesive subgroup for dichotomousrelationships is a subset of nodes among whom there are relatively strong, direct,
intense, frequent, or positive ties; and for valued relationships (like the ones encountered
in this work) the subsets of nodes will have strong orfrequent ties. In addition, these
authors identify four mechanisms by which these subgroups can be formed, namelymutuality of ties, closeness or reachability between members, frequency of ties
members, and relative frequency of ties among subgroup members compared to non-
members.
These authors contend that in the first three mechanisms, cohesive subgroups are
defined based on the relative strength, frequency, density, or closeness of ties within the
subgroups; whereas in the fourth this delineation is done on the relative weaknesses,
infrequency, sparseness or distance of ties from subgroup members to non members. In
a scale of strict to less strict, the forth mechanism (LS and Lambda Sets) is the mostrestrictive one, followed by the first and fifth one; and then by the other two. Table 2
summarizes these methods.
In this paper, we combine lambda sets and cliques as a first approach to measure
group cohesiveness because they allows to identify the Principal Investigators (or mostconnected people) in the network, the former, and all the (direct) collaborators of three or
more co-authors, the latter. Furthermore, the other two measures can be understood by
weakening the clique approach which can be further explored in future work.
Lambda sets
Lambda sets measure the level of cohesiveness of a group based on the frequency of tieswithin vs. outside subgroups. This type of subgroups are relatively robust in terms of itsconnectivity, i.e. it is difficult to disconnect by the removal of lines from the subgraph
(Wasserman and Faust, ibid). In order to assess the level of cohesiveness of a group this
algorithm varies the number of ties, d, within a group; the more ties you have to drop(within a group) the more cohesive this group is. In addition, this measure is used to
identify the most connected people within a network (something that we denote as PIs).
Table 3 Methods used to delimit subgroup
Mechanism Method Definition
Mutuality of
ties
cliques A clique is a maximal complete subgraph with at least three
nodes. It is a subset of nodes, all of which are adjacent to
each other, and there are no other nodes that are alsoadjacent to all of the members of a clique.
n-cliques An n-clique is a maximal subgraph in which the largest
geodesic distance between any two nodes is no greater thann. When n = 1, the subgraphs are cliques.
Reachability
and
diameter (or
closeness)
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
7/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
7
n-clans An n-clan is an n-clique in which the geodesic distance,
d(i,j), between all nodes in the subgraph is no greater than nfor paths within the subgraph.
n-clubs An n-club is a subgraph in which the distance between all
nodes within the subgraph is less than or equal to n;furthermore, no nodes can be added that also have geodesic
distance n or less from all members of the subgraph.
k-plexes A k-plexes is a maximal subgraph in which each node maybe lacking ties no more than ksubgraph members. When k
= 1, the subgraph is a clique.
Nodal
Degree(frequency of
ties
members) k-cores A k-cores is a subgraph in which each node is adjacent to at
least a minimum number, k, of the other nodes in thesubgraph.
LS Sets AnLSset is a subgroup definition that compares ties within
the subgroup to ties outside the subgroup by focusing on
the greater frequencies of ties among subgroup memberscompared to the ties between subgroup members to
outsiders.
Frequency
of ties within
vs. outside
subgroups
Lambda
Sets
A lambda setis a cohesive subset that is relatively robust in
terms of its connectivity, i.e. it is difficult to disconnect by
the removal of lines from the subgraph.
Source: Based on Wasserman and Faust (1994, pp. 251-267).
Cliques
A clique, consists of a subset of nodes, all of which are adjacent to each other, and there
are no other nodes that are also adjacent to all of the members of a clique (Wasserman
and Faust, 1994, pp. 254). As previously stated, this technique will allow us to identify allthe direct collaborators of a researcher.
3.2 Bibliometric Analysis
Since the 1970sperformance analysis (based on publication output and received citation)
(Martin and Irvine, 1983; van Raan, 2000; Kane, 2001; van Raan, 2005) and the mapping
of science (based on co-citation analysis) (Small, 1973; Narin, 1976; Narin, 1978; Small,1978; Leydesdorff, 1987; Gmr, 2003) have been widely used in evaluative bibliometrics
(see Noyons, Moed, and Luwel (1999) for a combined perspective; and van Raan, (2004)
for a historical development). However it is important to note that, with the exception of
Noyons, Moed, and Luwel (1999), these two approaches have been used separately. Inthe following section we briefly summarize both techniques.
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
8/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
8
Performance AnalysisThe traditional approach toward analysis of output performance by research units is to
use publications, citations, and patent counts as signals of work and productivity of the
focal unit. This method is based on the premise that an article will be published in a
refereed journal only when expert reviewers and the editor approve its quality and it willbe cited by other researchers as recognition of its authority (COSEPUP, 1999, p. 18). In
order to measure this output, several metrics have been developed in the literature [van
Raan, 2004; Thomson-ISI, 2003]. In this work, we use some of the most established
metrics, including total scientific output, citation counts and citation impact, defined as:
!" articlesPublishedoutputscientificTotal (1)
!" citationsReceivedcountCitation (2)
!!"
articlesPublished
citationsReceivedimpactCitation (3)
In addition, we also use a normalized citation impact, in order to have a measurement of
performance that is comparable across RGs, regardless of their size composition.
# $sizeGroupgroupperarticlesPublished
grouppercitationsReceivedimpactcitationNormalized %
%
&
'
((
)
*"
!!
(4)
Co-citation analysis
Co-citation analysis studies the structure and development of scientific communities andareas. This methodology is based on the notion that a citation is a valid and reliable
indicator of scientific communication, and this measure signals the relevance of an article
(Small, 1978; Garfield, 1979; Gmr, 2003). The specialized literature identifies a co-
citation if two publications or authors cite the same reference; and uses this concept as ameasure for similarity of content between the two publications or authors (Gmr, 2003).
In the last 30 years, two approaches have been developed within this framework,namely document co-citation (focused on documents and publications with peer-review
procedures) and author co-citation. In this work, we use the first approach to measure thesimilarity between the knowledge base (citations) of two RGs. For such purpose wedefine the knowledge footprint(KFP) for group i all the backward citations used by all
members of a group in all of their papers within a specific time frame.
4. Method
In this section we discuss the method developed for the identification of research groups
and for benchmarking their productivity and impact. This method consists of five steps.
First, the collaboration patterns among researchers will be pointed out, by identifying all
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
9/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
9
the direct and indirect ties of each researcher and measuring its Lambda Set level6
(LSL);
this collection of patterns and collaborators will form a set of collaboration groups (CG).Second, different research groups (RG) will be defined based on the researchers with the
highest LSL (identified in this work as PIs), the CGs of these authors, and characteristics
of their direct ties. Third, theKnowledge Footprint(KFP) of each group is delimited; and
a co-citation analysis is performed with this footprint to establish the similarities betweengroups. Fourth, the performance of each RG is measured using the absolute number of
articles and the number of articles per research group per researcher. Finally, the research
groups are benchmarked combining the previous results. Figure 1 shows a conceptual
representation of the method.
Figure 1 Method steps. This figure shows the steps of the method for the definition and
benchmark of research groups.
4.1 Identification of Collaboration Groups
In this step, the collaboration groups are established based on the pattern of co-authorshipin a field within a certain period of time. This step is subdivided into four sub-steps. First
(sub-step 1.1), all articles produced for a certain broad area of knowledge, time periodand relevant universe (e.g. country, region, city, university) are identified. This sub-step
generates a database containing a set of articles, authors, area of knowledge, backward
citations, and citations received within a certain timeframe.
Second (sub-step 1.2), a dichotomous mode-2 matrix (NxM, N Authors, M
articles) is created, allowing an author to be linked to all articles she or he has produced.
6 Labda Set level is the number of ties that need to be removed so an author is completely isolated
1. Identify co-authors
3. Define
research groups
6. Benchmark
research group
5. Measure
research group performance
4. Establish
knowledge footprint (KFP)
2. Define collaboration groups
Method Techinques
Bibliometric analysis
Perfromance analysis
Netowrk analysis
(grouping algorithms)
Co-citation analysis
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
10/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
10
Because co-authorship exists in many articles, this matrix also provides an indirect
relationship between authors (Figure A1 in appendix A shows an example of these typesof matrices).
Third (sub-step 1.3), to identify the direct relationships between authors, the
previous mode-2 matrix is converted into a mode-1 matrix (or adjacency matrix). This
produces a weightedNxN matrix (author by author) where the ijw values of each cell
indicate the strength or frequency of the relation (co-authorship) between authors i andj(Figure A2 in appendix A gives an example of this type of matrix).7
Fourth (sub-step 1.4), the most interconnected groups and people within a field ofknowledge are identified using the level of cohesiveness. Specifically, the concept of
Lambda Sets is used so by varying the number of ties (i.e. removing them) within a group
all relevant cohesive groups are found. Furthermore, this procedure provides the Lambda
Set Level (LSL) for each author, i.e. the number of ties that need to be removed so aresearcher becomes completely isolated; identifying the key players within a network, or
Principal Investigators (PIs). In addition, all collaboration groups (CG) within the area ofknowledge are identified using a second measure of cohesiveness. In this case the
concept ofclique is used so all the direct ties are found. For the purpose of the study, a
collaboration group will consist of three or more co-authors and/or authors that jointly
share 2 or more co-authors. In contrast, individual authors (authors that have co-authors
with no relation between) them will not be part of a collaboration group.
4.2 Research Group Delimitation
In this step we use the information on each author LSL, its CGs and the characteristics of
its direct ties to define the boundaries of the research groups (RG). The intuition of thisstep is that we first find all the Principal Investigators (PIs) within a particular knowledge
area, using the LSL of each author. After this, we assign each PI to a research group and,
based on the characteristics of its direct collaborators we allocate them to this or other
RGs. This entails four steps.
First (sub-step 2.1), the Principal Investigators within a particular knowledge area
are identify (by ordering and ranking all the authors based on their LSL) and assigned toa unique RG.
Second (sub-step 2.2), based on the CGs of the PI, all its direct collaborators areidentified and categorized as non-shared and shared collaborators. A direct collaborator is
defined as a non-shared collaborator if all its collaborations groups are a subset of the
collaborations groups of the PI. Conversely, a direct collaborator is defined as a shared
collaborator if its collaborations groups are part of collaborations groups of more thantwo PIs.
7 The conversion of a mode-2 matrix into mode-1 matrix, calculations fore extracting the cliques; as well as
the normalization and symmetrization of the KFP are performed using UCINET VI. This is a software has
built-in all the necessary for Network Analysis. In addition, we used NetDraw v. 2.25 to plot the differentnetworks, like figure YYY in section BBB.
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
11/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
11
Third (sub-step 2.3), the direct collaborators are immediately assign to a (PIs)
research group if they are categorized as non-shared collaborator. And for the reamingones a two stage rule is used. First, the overlap of CGs between the shared-collaborator
and the PIs is assed. The shared-collaborator is assigned to a particular RG if the CG
overlap between the PI of that group and the researcher is the highest. If the overlap were
the same, a second rule is used. For this case, the institutional affiliation of the shared-collaborator and the PIs RG is used, assigning this researcher to the group that has the
highest similarities in institutional affiliations.
Finally (sub-step 2.4), isolated dyads or individuals researchers with low LSL areassign to PIs (and its corresponding RG) with higher LSL using sub-step 2.3. Figure 2
provides an example for this part of the method (the identification of collaboration groups
and shared and non-shared collaborators).
Author AAuthor A
Research
Group for
Auhtor A
at a LSL
Research
Group for
Auhtor A
at a LSL
Non Shared
collaborators
Collaborators
of non-shared
collaborators
Collaborators
of non-shared
collaboratorsShared collaboratorsShared collaborators
Research
Group for
other auhtors
at other LSL
Research
Group for
other auhtors
at other LSL
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
12/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
12
Figure 2 Example of Research Group (RG) Delimitation. This figure provides and
example of how the proposed method delimits the boundaries of a RG. First, it starts withthe author (the PI of this group) that has the highest Lambda Set Level (in the case the
green square) and creates a group (yellow triangle). This process is repeated far all the
authors in a certain network. Second, all the collaborators of the author are linked to thisgroup and a distinction is made between shared (with other RG) and non-shared
collaborators. Third, the shared collaborators are assigned to this RG based on the LSL of
the PIs and if necessary on the institutional affiliation of the shared researcher, i.e. this
author will be assigned to the PIs RG with highest LSL and if necessary to the highestinstitutional overlap between the shared researcher and the RG
After taking these steps, the universe of analysis will be clustered in a set of RG,
as well as various researchers that are not integrated in any research group, as defined by
the method.
4.3 Knowledge footprint and group similarity
This part of the method is divided into four sub-steps. First (sub-step 3.1), we establish
the Knowledge Footprint (KFP) of each group. For such purpose, all backward citationsfor each RG are aggregated in a mode-2 matrix (KxJ,K research groups,J citations).
Second (sub-step 3.2), we convert this matrix into a KxK matrix using RG affiliation.
This produces a (directed) co-citation matrix (CCM) for all the research groups for a field
of knowledge within a certain period of time. Figure 3 shows a hypothetical example of
this type of matrix.
RG1 RG2 RG3 RG4 RG5 RG6 RG7 RG8 RG9 RG10
RG1 40 0 0 0 22 0 0 10 0 0
RG2 0 50 0 0 0 0 0 0 0 0
RG3 0 0 21 0 0 2 0 5 0 0
RG4 0 0 0 14 0 0 0 0 0 0
RG5 22 0 0 0 29 0 0 0 10 0
RG6 0 0 2 0 0 39 0 0 0 0
RG7 0 0 0 0 0 0 20 0 0 0
RG8 10 0 5 0 0 0 0 35 0 0
RG9 0 0 0 0 10 0 0 0 24 2
RG10 0 0 0 0 0 0 0 0 2 2 Figure 3 Co-citation matrix. This figure represents a hypothetical co-citation matrix.The value in the 11
thcell corresponds to the total number of backward citations for group
i. The value in the ijth
cell represents the total number of common citations between
groups i andj.
Third, the CCM is normalized with respect to the maximum number of citations
in each column. This produces a (directed) normalized co-citation matrix (NCCM).Figure 4 shows a hypothetical example of this type of matrix.
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
13/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
13
RG1 RG2 RG3 RG4 RG5 RG6 RG7 RG8 RG9 RG10
RG1 100% 0% 0% 0% 76% 0% 0% 29% 0% 0%
RG2 0% 100% 0% 0% 0% 0% 0% 0% 0% 0%
RG3 0% 0% 100% 0% 0% 5% 0% 14% 0% 0%RG4 0% 0% 0% 100% 0% 0% 0% 0% 0% 0%
RG5 55% 0% 0% 0% 100% 0% 0% 0% 42% 0%
RG6 0% 0% 10% 0% 0% 100% 0% 0% 0% 0%
RG7 0% 0% 0% 0% 0% 0% 100% 0% 0% 0%
RG8 25% 0% 24% 0% 0% 0% 0% 100% 0% 0%
RG9 0% 0% 0% 0% 35% 0% 0% 0% 100% 100%
RG10 0% 0% 0% 0% 0% 0% 0% 0% 8% 1 00% Figure 4NormalizedCo-citation matrix. This figure represents a hypothetical NCCM
matrix. The value of the ijth
cell represents the level of similarity between the group inthe i
throw with respect to the group in thej
thcolumn, e.g. the KFP of RG5 has a level of
similarity of 55% with respect to RG1. Conversely, RG1 has a 76% citation overlap with
respect to RG5.
4.4 Scientific output and performance
In order to measure the scientific performance of each research group, we sum thenumber of articles published by each research group, a proxy for RG scientific output,
and the citation counts, which is a proxy for RG total scientific impact. Finally, citation
impact by group size is established as a proxy for a normalized measure of scientificimpact.
4.5 Group benchmark
Once the research groups have been defined, the similarities among the RG establishedand their performance assessed, we proceed to benchmark groups against their peers. The
benchmark groups are identified through minimum thresholds in the level of similarity ofthe knowledge on which the groups rely for their work. We used a ten, five and one
percent overlap in KFP as the baseline levels for similarity between groups. Figures 8
shows a hypothetical research group, its peers (second column), the normalized co-
citation share (third column), the different levels of similarities (different scale of grays)
and in the bottom theKFP overlap.
The KFP is defined as:
!"CG
jjipKFP overla similarity (5)
where CG is the number of peer groups and j is a peer ofi. This indicator tells us howmuch the aggregated KFP of the peer group overlaps with group i.
5. Demonstration of the Method
5.1. Data
To test the proposed method, we use a database from Thomson Scientific (2004),
formerly known as the Institute for Scientific Information (ISI), owned by the MexicanCouncil for Science and Technology (CONACYT). This database contains all papers
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
14/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
14
published between 1980 and 2003 with at least one address in Mexico8. This database
contains the following information: article name, author(s), author(s), address(es), year of
publication, journal, volume, pages, backward citations (i.e. references) and total numberof citations received.
To illustrate the application of the method, we selected all the papers published in
Physics and Mathematics by five leading institutions (and the departments related tothese fields) in the period of 1997-1999. We chose this period because we wanted the
most recent publications, while allowing for a citation window of five years. (i.e. for the
papers published in 1997 we restricted the citation count to the period 1997-2001, for the
ones published in 1998 this window corresponded to the period 1998-2002, and for 1999we chose a 1999-2003 citation window). Table 4 provides a list of institutions and their
departments with the number of papers published in each of them, as well as the number
of forward citations. Finally, we provide an example of how to benchmark RG using the
method developed in this paper.
5.2 Main outcomes
Table 4 shows a traditional assessment done by institution (University or Research
Center). From this tables it can be seen that the Autonomous National University of
Mexico (UNAM) is the leading institution in absolute terms and ranks second if bothmeasures are normalized by the total number of researchers, while the National
Polytechnic Institute lag behind (among this selected institutions) in the critical output
and impact measures
Table 4 Number of Articles and Citations in Physics and Math for SelectedInstitutions, 1997-1999
Articles Citations
InstitutionTotal
(Rank)
Per
researcher
Total
(Rank)
Per
researcher
Autonomous National
University of Mexico
(UNAM)
1051
(1)2.2
3588
(1)7.4
Research and AdvancedStudies Centre
(CINVESTAV)
417 (2) 5.61145
(2)15.5
Metropolitan Autonomous
University Iztapalapa280 (3) 2.1 719 (3) 5.4
8 In the last stage of our analysis we realized that this database also contained publications with at least one
address in New Mexico and none in Mexico. We think that Thomson Scientific might have created this
database with the key wordMexico, including all the papers fromMexico andNew Mexico. We preserved
the papers with all the addresses because we did not want to mistakenly eliminate useful data, however ourprimary concern was to concentrate on Mexico.
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
15/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
15
(UAM-I)
Autonomous University of
Puebla (BUAP)202 (4) 2.4 499 (4) 5.9
National Polytechnic
Institute (IPN)114 (5) 0.4 218 (5) 0.8
In Table 5, each institution has been broken down by an administrative boundary.
These depend on the nature of the institution and may reflect, for example, a researchcenter focused just on research; an institute or department training graduate students and
doing research; or a school primarily training undergrad students and doing some
research. This Table reflects the typical level of detail allowed by existing methods. It
already provides a more complete perspective than a purely institutional characterization.For example, the Physics Institute in UNAM has a stellar performance (in total output),
stronger than the Faculty of Science and much different from the Applied Math andSystems Research Institute. By contrast, in the Autonomous University of Puebla (fourth
place at the institutional level) the Physics Institute performs above two Departments ofthe Research and Advanced Studies Centre (second place at the institutional level).
However, if we would now look at the patterns of co-authorship and collaborationand identify research groups based on these patterns, a different and much more complete
perspective emerges. Figure 6 shows the distribution of research groups, identified by the
method, ranked by the number of publications published by each research group. Fromthis picture it can be seen that the performance of RG is more heterogeneous, e.g. UNAM
lead both top (for good) and bottom percentiles (for bad) with eight and eleven RGs,
respectively. Whereas the RG form BUAP and IPN trail the other institutions.
Table 5 Total Number of Articles and Citations in Physics and Math by Department
for Selected Institutions, 1997-1999
Department
Number
of
Articles
(Rank)
Number
of
Citations
(Rank)
Autonomous National University of MexicoPhysics Institute (IF UNAM ) 594 (1) 2495 (1)
Faculty of Science (FC UNAM ) 285 (3) 709 (3)Mathematics Institute (MI UNAM) 114 (6) 271 (6)Applied Math and Systems Research Institute
(IIMAS)82 (8) 212 (8)
Research and Advanced Studies Centre
Physics Department (Phy CINVESTAV) 289 (2) 879 (2)Applied Physics Department (APhy
CINVESTAV)83 (8) 204 (9)
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
16/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
16
Math Department (Math CINVESTAV) 55 (9) 88 (12)
Metropolitan Autonomous University
Iztapalapa
Physics Department (Phy UAM-I) 225 (4) 625 (4)
Math Department (Math UAM-I) 55 (9) 94 (11)
Autonomous University of Puebla
Physics Institute (IF BUAP) 124 (5) 356 (5)Faculty of Science (FC BUAP) 96 (7) 171 (10)
National Polytechnic Institute
School of Physical and Mathematics (ESFM) 114 (6) 218 (7)
Furthermore, if we normalized this distribution by the number of researchers ineach group a more complex picture emerges. Figure 6 shows the distribution of research
groups, identified by the method, ranked by the number of publications published by eachresearch group and normalized by group size. From this picture it can be seen that the
performance of RG is quite heterogeneous across institutions, UNAM doesnt lead thetop percentile and still leads the bottom one, and the number of RG in BUAP, IPN and
UAM are about the same.
In Figure 8, each institution has been broken down by an administrative
boundary, showing the distribution of RG (identified by the method) by department
ranked by the number of papers per RG. Once again it can be seen that the performanceof research groups is quite heterogeneous across departments. Furthermore, the same
heterogonous pictures emerge if the groups are ranked by number of citations (figures A3
and A4 in the appendix).
Figure 6 DistributionResearch Groups (RG) by Institution, Ranked by Number ofPapers per RG. This figure provides the distribution of RG based on the total number of
0
2
4
6
8
10
12
14
16
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Percentile
Numberofresearchgroups
BUAP
IPN
UNAM
UAM-I
CINVESTAV
0
2
4
6
8
10
12
14
16
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
0
2
4
6
8
10
12
14
16
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Percentile
Numberofresearchgroups
BUAP
IPN
UNAM
UAM-I
CINVESTAV
BUAPBUAP
IPNIPN
UNAMUNAM
UAM-IUAM-I
CINVESTAVCINVESTAV
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
17/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
17
articles produced by RG within a certain institution. The left hand side shows theinstitutions that have the highest number of leading groups., e.g. UNAM has eight RG in
the top percentile, where as BUAP has three. In contrast, the right side gives the number
of laggard research groups, e.g. UNAM has 11 and BUAP has only one. From this figureit can be seen that the performance of RG is more heterogeneous across institutions.
Figure 7 DistributionResearch Groups (RG) by Institution, Ranked by Number ofPapers per RG per researcher. This figure provides the distribution of RG based on thetotal number of articles produced by RG per researcher within a certain institution. The
left hand side of this figure, shows the top performing institution by number of researchgroups in the top percentile, e.g. UNAM has only one RG in the top percentile, where as
BUAP has four and UAM-I has the highest number of RG, eight. In contrast, the rightside gives the least performing institution, e.g. UNAM has 17, BUAP three and IPN
none. From this figure it can be seen that the performance of RG is quite heterogeneous
across institutions and the overall performance of each institution change if the size of theRG is taken into account.
0
2
4
6
8
10
12
14
16
18
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
0
2
4
6
8
10
12
14
16
18
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Percentile
Numbero
fresearchgroups
BUAP
IPN
UNAM
UAM-I
CINVESTAV
BUAPBUAP
IPNIPN
UNAMUNAM
UAM-IUAM-I
CINVESTAVCINVESTAV
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
18/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
18
Figure 8 DistributionResearch Groups (RG) by Department, Ranked by Number ofPapers per RG. This figure provides the distribution of RG based on the total number ofarticles produced by RG within a certain department. The left hand side shows the
departments that have the highest number of leading groups. And the right side gives the
number of laggard research groups by department. From this figure it can be seen that theperformance of RG is quite heterogeneous across departments.
As suggested this analysis so far, identifying research groups should allow a much better understanding of the dynamics of scientific productivity within and across
institutions and departments. Yet, this process also suggests that these groups will havevery different characteristics in what concerns their research nature. Therefore, it may not
be reasonable to compare groups across the board, but rather complement the ideas
described above with a method that would allow an identification of other benchmark
groups with a comparable research profile. As described in detail below, the methodallow us to compare a given group with others that produce comparable research by
looking at the overlap of citations in the papers they publish. Table 6 provides an
example for RG 080, overall this group ranks 32 out of 143 (for publications by research
group by researcher) and 24 (for citations by research group by researcher). However,
within its cohort ranks 5th and 4th respectively and 1st at a level of similarity of 5% ormore.
Furthermore, if we break the RG by administrative boundary we can see that this
group is located in the 20th
percentile, vs. the 80th
percentile if this group is justbenchmarked in the overall network (Figure 8).
The example above shows the power of combining a method than allows anidentification of research group boundaries with another that characterizes distances in
their knowledge footprint. The potential is for a very sharp and clear identification and
0
2
4
6
8
10
12
14
16
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
0
2
4
6
8
10
12
14
16
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
APhyCINVESTAV
PhyCINVESTAV
PhyUAM-IMathCINVESTAV
MathUAM-I
ESFM
FCUNAM
FCBUAP
IIMAS
IF-BUAP
IF-UNAM
MI-UNAM
APhyCINVESTAVAPhyCINVESTAV
PhyCINVESTAVPhyCINVESTAV
PhyUAM-IPhyUAM-IMathCINVESTAVMathCINVESTAV
MathUAM-IMathUAM-I
ESFMESFM
FCUNAMFCUNAM
FCBUAPFCBUAP
IIMASIIMAS
IF-BUAPIF-BUAP
IF-UNAMIF-UNAM
MI-UNAMMI-UNAM
Percentile
Numberofresearchgroups
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
19/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
19
benchmark of the relevant pockets of knowledge generation and impact in an institution
or region, allowing a more precise evaluation and reward process for university
administrators, program managers or policy makers.
Table 6 Relative Ranking Based on Number of Articles per Researcher
Using Group Knowledge Similarity Over 3 Years
Articles Citations
Ranking RankingResearch
Group
Level of
Similarity Per
ResearcherWithin
cohortOverall
Per
ResearcherWithin
cohortOverall
52 2,50% 2,5 1 17 3,5 1 16
017_2 16,80% 1,7 2 32 2,8 2 22
30 16,80% 1,7 2 32 2,7 3 24
=> 80 BASE 1,7 2 32 2,7 3 24047_1 16,80% 1,3 3 40 1,4 5 47
89 2,50% 1,3 3 39 0 12 40
013_12 14,90% 1,1 4 49 1,7 4 38
76 4,30% 1,1 4 50 1,1 8 57012_1 7,50% 1 5 52 1 9 59
022_4 7,50% 0,8 6 58 1,3 6 52
049_2 16,10% 0,8 6 55 1 9 59018_6 23,60% 0,8 6 58 1 9 59
92 87,60% 0,7 7 63 1,2 7 53
026_3 23,60% 0,7 7 62 1,1 8 57024_5 23,60% 0,7 7 64 0,9 10 61
007_6 1,20% 0,7 7 64 0,8 11 62
118 17,40% 0,6 8 67 0,8 11 64
Overall group 80 ranks 32 out of 143, but within its cohort (17 RG at a 1% level ofsimilarity) it ranks 2
nd.
6. Discussion and Policy Implications
In the last thirty years the realm of science and technology has evolved dramatically.These changes have spurred an evaluation culture (and industry), enhancing traditional
methods (like peer review (Guston, 2003)) or creating new ones, including benchmarking
between countries (May, 1997; Adams, 1998; King, 2004; Veloso, Reyes-Gonzalez andGonzalez-Brambila, 2005) or socioeconomic assessments (van Raan, 2000). While
useful, these evaluations methods have an important common limitation: the boundaries
of the focal unit are typically artificial and rigid, failing to notice unique and self
organizing characteristics of the research endeavor.
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
20/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
20
To address this limitation, this paper develops an evaluation method that takes
into account the endogenous, or self organizing, characteristics of research group. It
defines research groups based on the strength and frequency of the collaboration patterns(within a field of knowledge) and ranks them using the level of similarity of their
knowledge footprint (i.e. common citations). In addition, this method is tested with a
database from the fields of Mathematics and Physics in Mexico containing all the papers
published between 1997 and 1999 by 5 leading institutions (as reported by ISI); and adetailed full and relative peer benchmark is performed for both areas.
The method developed in this paper produced two main results. First, as expected,
the strength and frequency of the collaboration patterns allowed us to single out cohesivegroups, i.e. this method identifies the key research groups (or collective actors) in a field
of knowledge regardless of the institutional or location context of the members
(researchers). This is a departure from traditional methods because, under this approach,
a potential evaluator will not be able to assess the internal cohesiveness of groups, or sefl-organizing mechanisms. In addition, this method can be used to breakdown groups that
extend from the traditional boundaries, i.e. a multi-institutional group could be divided intwo.
Second, the knowledge footprint (KFP) and the benchmark at different levels of
similarity in KFP is an important departure from the established evaluation literature.This step allows potential evaluators to identify similar research groups, assess these
groups and produce more meaningful comparisons and rankings (e.g. see table 8). This
solution contrasts with the more traditional approach, where the evaluator typically uses
broad and artificial similarities, such as comparing mechanical engineering departments
across universities, assuming that they are more or less the same. In addition, this methodhas an important feature: it can easily be extended to other type of focal units, including,institutions, departments, networks or regions, and even individuals, with minor
modifications.
From the preliminary results it can be concluded that this method can support
policy makers and scientist to better identify the frontier of research groups and to find
suitable and relevant peers for benchmark. In addition, this procedure allows scholars, as
well as policy makers, to better understand the self-organizing mechanisms of researchgroups and assess how they evolve over time. We believe this whole process will
increased our knowledge of the research endeavor and combined with other methods (like
peer review) will produce better assessments. In addition, this method helps to close thegap between performance analysis and the mapping of science in bibliometric analysis,
by giving a different perspective to Noyons, Moed, and Luwel (1999). It also creates alink between the areas of research evaluation and network analysis.
But the development of this new method also generates new questions that needto be addressed in subsequent work. One of them is the effect of weakening the clique
assumption, using other measures of group cohesiveness (e.g. n-cliques, k-plexes, etc.) to
define collaborative groups. Another issue is to understand the impact of aggregatinggroups with a low Lambda Set Level (LSL) into groups with a higher LSL.. In addition,
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
21/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
21
further analysis is also need to test the robustness of this method, by incorporating other
fields knowledge or using data from other countries or regions.
Finally this method could further our understanding of the determinants of
research group productivity (Gonzalez-Brambila, Veloso and Krackhardt, 2005) in a
number of ways. One possibility is to study how the characteristics of the naturally
emerging groups are tied to their productivity. Another possibility would be to extend theapproach to other type of research output data amenable to equivalent analysis, in
particular patents.
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
22/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
22
ReferencesAdams, J., 1998. Benchmarking International Research. Nature 396, 615-618.
Carley, K. M. 2002. Smart Agents and Organizations of the Future. The Handbook of
New Media. Edited by Leah Lievrouw & Sonia Livingstone, Ch. 12 pp. 206-220,
Thousand Oaks, CA, Sage.
Carley, K. M., 2003. Dynamic Network Analysis. Dynamic Social Network Modeling
and Analysis: Workshop Summary and Papers, Eds. Ronald Breiger, Kathleen
Carley, and Philippa Pattison, Committee on Human Factors, National Research
Council, National Research Council. 133-145.
COSEPUP. 1999. Evaluating Federal Research Programs: Research and the Government
Performance and Results Act. National Academy Press, Washington, DC.
Frederiksen, L. F., Hansson, F. and Wenneberg, S. B., 2003. The Agora and the Role of
Research. Evaluation 9, 149172.
Garfield, E., 1979. Is Citation Analysis a Legitimate Evaluation Tool? Scientometrics 1,
359-375.
Georghiou, L. and Roessner, D., 2000. Evaluating technology programs: tools and
methods. Research Policy 29, 657678.
Glser, J., Spurling T. H. and Butler, L., Intraorganisational evaluation: are there least
evaluable units? Research Evaluation 13, 19-32.
Gmr, M., 2003. Co-citation analysis and the search for invisible colleges: A
methodological evaluation. Scientometrics 57, 27 57.
Gonzalez-Brambila, C. N., Veloso, F., and Krackhardt, D., 2005. Social Capital and the
Creation of Knowledge. Working Paper. Carnegie Mellon University, Pittsburgh, PA.
Guston, D. H., 2003. The expanding role of peer review process in the United States. In:Shapira, P., Kuhlmann, S. (Eds.), Learning from Science and Technology Policy
Evaluation. Edward Elgar Publishing, Northampton, MA, p. 81-97
Guimer, R., Uzzi, B., Spiro, J. and Nunes Amaral, L. A. 2005. Team Assembly
Mechanisms Determine Collaboration Network Structure and Team Performance.Science 308, 697-702.
Kane, A., 2001. Indicators and Evaluation for Science, Technology and Innovation. Paperfor an ICSTI working groups on evaluating basic research. Retrieved from
http://www.economics.nuigalway.ie/people/kane/2002_2003ec378.html; August
2005.
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
23/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
23
King, A. D., 2004. The Scientific Impact of Nations. Nature 430, 311-316.
Krackhardt, D., and Carley, K. M., 1998. A PCANS Model of Structure in Organization.Proceedings of the 1998 International Symposium on Command and Control
Research and Technology. Conference held in June. Monterray, CA. Evidence Based
Research, Vienna, VA, 113-119.
Leydesdorff, L., 2005. The Evaluation of Research and the Evolution of Science
Indicators. Current Science 89, 1510-1517.
Leydesdorff, L., 1987. Various methods for the Mapping of Science. Scientometrics 11,291-320.
Martin, B. R. and Irvine, J., 1983. Assessing basic research: Some partial indicators of
scientific progress in radio astronomy. Research Policy 12, 61-90.
May, R. M., 1997. The Scientific Wealth of Nations. Science 275, 793796.
Narin, F., 1976. Evaluative Bibliometrics: The Use of Publication and Citation Analysis
in the Evaluation of Scientific Activity. National Science Foundation, Washington
D.C.
Narin, F., 1978. Objectivity versus relevance in studies of scientific advance.
Scientometrics 1, 35-41.
Noyons, E.C.M., Moed, H.F. and Luwel, M., 1999. Combining Mapping and CitationAnalysis for Evaluative Bibliometric Purposes: A Bibliometric Study. Journal of theAmerican Society for Information Science 50, 115131.
Papaconstantinou, G. and Polt, W., 1997. Policy Evaluation and Technology: AnOverview. OECD Proceedings, Conference on Policy Evaluation in Innovation and
Technology. OECD Paris.
van Raan, A. F. J., 2000. R&D evaluation at the beginning of the new century. ResearchEvaluation 9, 81-86.
van Raan, A. F. J., 2004. Measuring Science Capita Selecta of Current Main Issues. In:Moed, H.F., Glnzel, W., and Schmoch, U. (Eds.), Handbook of Quantitative Science
and Technology Research. Kluwer Academic Publishers, Dordrecht, p.19-50.
van Raan, A. F. J., 2005. Fatal Attraction: Conceptual and methodological problems inthe ranking of universities by bibliometric methods. Scientometrics 62, p. 133-143.
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
24/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
24
Rip. A., 2000. Societal Challenges for R&D Evaluation. Proceedings from US-EU
Workshop Learning from Science and Technology Policy Evaluation. (Eds.) Shapira,
P., Kuhlmann, S. Bad Herrenalb, Germany, September 2000.
Rogers, J. D., Bozeman, B., and Chompalov, I., 2001. Obstacles and opportunities in the
application of network analysis to the evaluation of R&D. Research Evaluation 10,
161-172.
Small, H., 1973. Co-citation in scientific literature: A new measure of the relationship
between publications. Journal of the American Society for Information Science 24,
265269.
Small, H. G., 1978. Cited documents as concept symbols. Social Studies of Science 8,
327-340.
Thomson-ISI, 2003. National Science Indicators database Deluxe version. Thomson
Research, Philadelphia, PA.
Thomson Scientific, 2004. National Citation Report 1980-2003: customized database for
the Mexican Council for Science and Technology (CONACYT). Thomson Research,
Philadelphia, PA.
Veloso, F., Reyes-Gonzalez, L. and Gonzalez-Brambila, C. N., 2005. The Scientific
Impact of Developing Nations. Working Paper. Carnegie Mellon University,
Pittsburgh, PA.
Wasserman, S. and Faust, K., 1994. Social Network Analysis: Methods and Applications.Cambridge University Press, Cambridge, UK.
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
25/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
25
Appendix A
Figures A1 and A2 are used in de description of the method provide. Figure A1 provides
an example of a dichotomous mode-2 matrix (NxM, N Authors, M articles). And
figure A2 shows mode-1 matrix (or adjacency matrix) of author by author.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
1 ABREU, CRA 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2 AGUDO, AL 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
3 AGUILAR, R 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
4 AGUILAR-RIOS, G 0 0 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
5 AGUILAR-RODRIGUEZ, E 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 0 0
6 AGUILARA, R 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
7 AHONEN, P 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
8 ALVARADO, JFJ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
9 ALVAREZ, J 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
10 ALVAREZ-RAMIREZ, J 0 0 0 0 1 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0
11 ANCHEYTA-JUAREZ, J 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 0 0
Articles
Author
Figure A1 Hypothetical example of a mode-2matrix (author by paper). In this case
each row makes reference to all the articles published by a particular author and each
column relates all the co-authors of an article. A 1 in the ijth
cell indicates that authori
is related to publicationj. A 0 indicates no relationship.
1 2 3 4 5 6 7 8 9 10 11
AB AG AG AG AG AG AH AL AL AL AN1 ABREU, CRA 1 0 0 0 0 0 0 0 0 0 0
2 AGUDO, AL 0 2 0 0 0 0 0 0 0 0 0
3 AGUILAR, R 0 0 7 0 0 0 0 0 0 1 0
4 AGUILAR-RIOS, G 0 0 0 2 0 0 0 0 0 0 0
5 AGUILAR-RODRIGUEZ, E 0 0 0 0 5 0 0 0 0 0 5
6 AGUILARA, R 0 0 0 0 0 1 0 0 0 1 0
7 AHONEN, P 0 0 0 0 0 0 1 0 0 0 0
8 ALVARADO, JFJ 0 0 0 0 0 0 0 1 0 0 0
9 ALVAREZ, J 0 0 0 0 0 0 0 0 3 0 0
10 ALVAREZ-RAMIREZ, J 0 0 1 0 0 1 0 0 0 18 0
11 ANCHEYTA-JUAREZ, J 0 0 0 0 5 0 0 0 0 0 5
Author
Author
Figure A2Hypothetical example of adjacency matrix (author by author). This matrix
relates an author with all its co-authors. The value in the ijth
cell indicates the strength of
frequency between authori andj
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
26/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
26
Table A1 gives number of research groups by Institution and Department identified by
the proposed method
Table A1 Total number of research groups in Physics and Math for selected
institutions, 1997-1999
Number of Research Groups
Institution and Departmentby
Institution*
by
Department*
Autonomous National University of Mexico
(UNAM)104
Physics Institute (IF UNAM ) 99Faculty of Science (FC UNAM ) 27
Mathematics Institute (MI UNAM) 7
Applied Math and Systems Research Institute
(IIMAS)6
Research and Advanced Studies Centre
(CINVESTAV IPN)61
Physics Department (Phy CINVESTAV) 46Applied Physics Department (APhy
CINVESTAV)23
Math Department (Math CINVESTAV) 6
Metropolitan Autonomous University Iztapalapa
(UAM-I)44
Physics Department (Phy UAM-I) 42
Math Department (Math UAM-I) 3
Autonomous University of Puebla (BUAP) 33
Physics Institute (IF BUAP) 24
Faculty of Science (FC BUAP) 16
National Polytechnic Institute (IPN) 31School of Physical and Mathematics (ESFM) 31
TOTAL number of RG 143
*There is an overlap of RG across Institutions and Departments so the sum in more than
the total number of RG (143)
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
27/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
27
Figure A3 shoes the distribution RG by institution ranked by number of citations per RG
(top) and by number of citations per RG per researcher (bottom). And figure A4 shows
the distribution RG by department ranked by number of citations per RG.
Figure A3 Distribution Research Groups (RG) by Institution, Ranked by Number of
Citations per RG (TOP) and by Number of Citations per RG per researcher
(BOTTOM). This figure provides the distribution of RG based on the total number of
citations produced by RG and the normalization of this measure controlling by groupsize. The left hand side of this figure, shows the top performing RG by institution and the
right side gives the least performing RG. From this figure it can be seen that the
0
2
4
6
8
10
12
14
16
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
0
2
4
6
8
10
12
14
16
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Distribution Research Groups (RG) by Institution
Ranked by Number of Citation per RG per researcher
Percentile
Numberofresearchgroups
Distribution Research Groups (RG) by Institution
Ranked by Number of Citation per RG
0
2
4
6
8
10
12
14
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
0
2
4
6
8
10
12
14
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Percentile
Numberofre
searchgroups
BUAP
IPN
UNAM
UAM-I
CINVESTAV
BUAPBUAP
IPNIPN
UNAMUNAM
UAM-IUAM-I
CINVESTAVCINVESTAV
-
8/2/2019 Leonardo Reyes Gonzalez Evaluation of Mexican Groups
28/28
Paper presented in the Prime-Latin America Conference at Mexico City, September 24-26 2008
28
performance of RG is quite heterogeneous across institutions and the overall performance
of each institution change if the size of the RG is taken into account.
Figure A4 DistributionResearch Groups (RG) by Department, Ranked by Numberof Citations per RG. This figure provides the distribution of RG based on the total
number of citations produced by RG within a certain department. The left hand side
shows the departments that have the highest number of leading groups. And the right side
gives the number of laggard research groups by department. From this figure it can beseen that the performance of RG is quite heterogeneous across departments.
0
2
4
6
8
10
12
14
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
0
2
4
6
8
10
12
14
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
APhyCINVESTAV
PhyCINVESTAV
PhyUAM-I
MathCINVESTAV
MathUAM-I
ESFM
FCUNAM
FCBUAP
IIMAS
IF-BUAP
IF-UNAMMI-UNAM
APhyCINVESTAVAPhyCINVESTAV
PhyCINVESTAVPhyCINVESTAV
PhyUAM-IPhyUAM-I
MathCINVESTAVMathCINVESTAV
MathUAM-IMathUAM-I
ESFMESFM
FCUNAMFCUNAM
FCBUAPFCBUAP
IIMASIIMAS
IF-BUAPIF-BUAP
IF-UNAMIF-UNAMMI-UNAMMI-UNAM
Percentile
Numberofresearchgroups