the e ect on citation inequality of di erences in citation ... · the e ect on citation inequality...
TRANSCRIPT
![Page 1: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/1.jpg)
The E�ect on Citation Inequality of Di�erences in
Citation Practices across Scienti�c Fields
Juan A. Crespo1, Yunrong Li2, Javier Ruiz-Castillo2
2Universidad Carlos III de Madrid, Spain
October 10, 2013
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 1 / 19
![Page 2: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/2.jpg)
Outline
1 Motivation
2 Model
3 Empirical with Raw Data
4 Normalization of Raw Data
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 2 / 19
![Page 3: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/3.jpg)
Motivation
How to assess the scienti�c in�uence of a research paper?
Citation impact: the number of citations received by the paper within acertain period of time after its publication.
A Scienti�c Field: a collection of papers published in a set of closelyrelated professional journals.
Empirical Regularity: highly skewed citation distributions(Albarrán,2011), citation inequality is very large within a �eld as wellas in all-�elds case.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 3 / 19
![Page 4: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/4.jpg)
Skewness of Citation Distribution
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 4 / 19
![Page 5: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/5.jpg)
Motivation
The large citation inequality may be due to di�erent papers havedi�erent scienti�c in�uence, or papers belong to di�erent �elds.
The �eld dependence of citation impacts:
Size of the �eld: average number of papers per author in a givenperiod of time.
Average number of references per paper.
The speed at which the citation process evolves.
To introduce a simple model to see how important is the �elddependence of citation impacts on citation inequality.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 5 / 19
![Page 6: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/6.jpg)
Motivation
The large citation inequality may be due to di�erent papers havedi�erent scienti�c in�uence, or papers belong to di�erent �elds.
The �eld dependence of citation impacts:
Size of the �eld: average number of papers per author in a givenperiod of time.
Average number of references per paper.
The speed at which the citation process evolves.
To introduce a simple model to see how important is the �elddependence of citation impacts on citation inequality.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 5 / 19
![Page 7: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/7.jpg)
Motivation
The large citation inequality may be due to di�erent papers havedi�erent scienti�c in�uence, or papers belong to di�erent �elds.
The �eld dependence of citation impacts:
Size of the �eld: average number of papers per author in a givenperiod of time.
Average number of references per paper.
The speed at which the citation process evolves.
To introduce a simple model to see how important is the �elddependence of citation impacts on citation inequality.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 5 / 19
![Page 8: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/8.jpg)
Statistics of Field Citation Distribution
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 6 / 19
![Page 9: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/9.jpg)
Roemer´s Income Inequality Model
Roemer´s (1998) model:
Income of individuals depends on e�orts and circumstances (e.g.parents´ wealth, education. ), partition the population by �type�,
Incometi = (typet , effortti ).
E�ort distribution within a type is a characteristic of the type.
A1: Within a type, individuals at the same quantile of e�ortdistribution implement the same �degree� of e�ort.
A2: Within a type, income is monotonic in e�ort. Quantiles of e�ortdistribution correspond to quantiles of income distribution.
Holding constant the degree of e�ort/income, income inequality acrosstypes is due to circumstances, �inequality of opportunity�.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 7 / 19
![Page 10: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/10.jpg)
Analogy of Our Model with Roemer´s
Individuals=⇒Articles
Income=⇒Citation impact
A income distribution=⇒A citation distribution
Circumstances/Types=⇒Fields
E�ort=⇒Scienti�c in�uence
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 8 / 19
![Page 11: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/11.jpg)
Assumptions
Citationfi = (fieldf , Scientific Influencefi ), f = 1, ...,F ; i = 1, ...,N.
A1: Within a �eld, articles at the same quantile of the scienti�cin�uence distribution re�ect the same �degree� of scienti�c in�uence.
A2: Citationfi is monotonic in scientific influencefi .
Quantiles of scienti�c in�uence distribution correspond the quantiles ofcitation distribution.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 9 / 19
![Page 12: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/12.jpg)
Assumptions
Citationfi = (fieldf , Scientific Influencefi ), f = 1, ...,F ; i = 1, ...,N.
A1: Within a �eld, articles at the same quantile of the scienti�cin�uence distribution re�ect the same �degree� of scienti�c in�uence.
A2: Citationfi is monotonic in scientific influencefi .
Quantiles of scienti�c in�uence distribution correspond the quantiles ofcitation distribution.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 9 / 19
![Page 13: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/13.jpg)
Assumptions
Citationfi = (fieldf , Scientific Influencefi ), f = 1, ...,F ; i = 1, ...,N.
A1: Within a �eld, articles at the same quantile of the scienti�cin�uence distribution re�ect the same �degree� of scienti�c in�uence.
A2: Citationfi is monotonic in scientific influencefi .
Quantiles of scienti�c in�uence distribution correspond the quantiles ofcitation distribution.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 9 / 19
![Page 14: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/14.jpg)
Double Partition
Partition the all-�elds citation distribution by �elds f and quantiles π.
Field 11st
2nd...
Πth
quantiles
...
Field F1st
2nd...
Πth
quantiles
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 10 / 19
![Page 15: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/15.jpg)
Double Partition
�sort�, citations are in asending order within each �eld.
�_pctile�, create 1000 quantiles within each �eld.
�merge� to merge all �elds together.
All-�elds citation distribution is a matrix of cells, (f , π).
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 11 / 19
![Page 16: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/16.jpg)
Decomposable Inequality Index
Generalized Entropy (GE) family of inequality indices, the Theil index.
I1(C ) = 1
N ∑lclµ
log clµ
, citation inequality of all-�elds case, µ is the meancitation of all articles.
For articles with 0 ciation, we follow the convention 0∗ log0 = 0.
I1(C ) is decomposable, I1(C ) = W + S + IDCP.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 12 / 19
![Page 17: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/17.jpg)
Decomposable Inequality Index
Generalized Entropy (GE) family of inequality indices, the Theil index.
I1(C ) = 1
N ∑lclµ
log clµ
, citation inequality of all-�elds case, µ is the meancitation of all articles.
For articles with 0 ciation, we follow the convention 0∗ log0 = 0.
I1(C ) is decomposable, I1(C ) = W + S + IDCP.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 12 / 19
![Page 18: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/18.jpg)
Decomposable Inequality Index
Generalized Entropy (GE) family of inequality indices, the Theil index.
I1(C ) = 1
N ∑lclµ
log clµ
, citation inequality of all-�elds case, µ is the meancitation of all articles.
For articles with 0 ciation, we follow the convention 0∗ log0 = 0.
I1(C ) is decomposable, I1(C ) = W + S + IDCP.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 12 / 19
![Page 19: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/19.jpg)
Decomposable Inequality Index
W = ∑π ∑f νπ,f I1(cπ
f ), Within-Group term.
Weight:νπ,f is share of total citations received by articles in cell (f , π).For large Π, W is small.
S = I1(µ1, ...µπ , ...µΠ), Between-Group term.
Each paper is given the mean citation of articles in its own quantile.Citation inequality is due to articles belonging to di�erent quantiles, i.e.skewness of the all-�elds citation distribution.
IDCP = ∑π νπ I1(µπ1 , ...µ
π
f , ...µπ
F ).
Each paper is given the mean citation of articles in its own cell. Withinany quantile π, citation inequality is due to the di�erences in citationpractices across �elds.Weight: νπ is the share of total citations received by articles inquantile π.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 13 / 19
![Page 20: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/20.jpg)
Decomposable Inequality Index
W = ∑π ∑f νπ,f I1(cπ
f ), Within-Group term.
Weight:νπ,f is share of total citations received by articles in cell (f , π).For large Π, W is small.
S = I1(µ1, ...µπ , ...µΠ), Between-Group term.
Each paper is given the mean citation of articles in its own quantile.Citation inequality is due to articles belonging to di�erent quantiles, i.e.skewness of the all-�elds citation distribution.
IDCP = ∑π νπ I1(µπ1 , ...µ
π
f , ...µπ
F ).
Each paper is given the mean citation of articles in its own cell. Withinany quantile π, citation inequality is due to the di�erences in citationpractices across �elds.Weight: νπ is the share of total citations received by articles inquantile π.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 13 / 19
![Page 21: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/21.jpg)
Decomposable Inequality Index
W = ∑π ∑f νπ,f I1(cπ
f ), Within-Group term.
Weight:νπ,f is share of total citations received by articles in cell (f , π).For large Π, W is small.
S = I1(µ1, ...µπ , ...µΠ), Between-Group term.
Each paper is given the mean citation of articles in its own quantile.Citation inequality is due to articles belonging to di�erent quantiles, i.e.skewness of the all-�elds citation distribution.
IDCP = ∑π νπ I1(µπ1 , ...µ
π
f , ...µπ
F ).
Each paper is given the mean citation of articles in its own cell. Withinany quantile π, citation inequality is due to the di�erences in citationpractices across �elds.Weight: νπ is the share of total citations received by articles inquantile π.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 13 / 19
![Page 22: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/22.jpg)
Data
Only research papers. One paper is assigned into only one �eld.
About 4.4 million articles pubilshed in 1998-2003.
A common �ve-year citation window for every year, about 35 millioncitations.
22 broad �elds: 20 for natural sciences and 2 for social sciences,distinguished by Thomson Scienti�c.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 14 / 19
![Page 23: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/23.jpg)
Results with Raw Data
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 15 / 19
![Page 24: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/24.jpg)
Exchange Rates
De�ne �exchange rate�, ef (π), for articles in cell (f , π):
ef (π) =µπ
fµπ , how many citations for an article at quantile π of �eld f
are equivalent on average to one citation in all-�elds case (thereference situation).
If ef (π) varies dramatically across π within a �eld, no common factorof all quantiles can be estimated.
Empirically, ef (π) remain su�ciently constant over quantiles[706th, 998th]. 60-70% of citations in each �eld.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 16 / 19
![Page 25: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/25.jpg)
Exchange Rates
De�ne �exchange rate�, ef (π), for articles in cell (f , π):
ef (π) =µπ
fµπ , how many citations for an article at quantile π of �eld f
are equivalent on average to one citation in all-�elds case (thereference situation).
If ef (π) varies dramatically across π within a �eld, no common factorof all quantiles can be estimated.
Empirically, ef (π) remain su�ciently constant over quantiles[706th, 998th]. 60-70% of citations in each �eld.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 16 / 19
![Page 26: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/26.jpg)
Exchange Rates
De�ne �exchange rate�, ef (π), for articles in cell (f , π):
ef (π) =µπ
fµπ , how many citations for an article at quantile π of �eld f
are equivalent on average to one citation in all-�elds case (thereference situation).
If ef (π) varies dramatically across π within a �eld, no common factorof all quantiles can be estimated.
Empirically, ef (π) remain su�ciently constant over quantiles[706th, 998th]. 60-70% of citations in each �eld.
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 16 / 19
![Page 27: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/27.jpg)
Normalization Factor
De�ne an average-based exchange rate (ER) over [706th, 998th]:
ef = 1πM−πm
∑π ef (π)
Normalize raw citations: c∗fi = cfief
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 17 / 19
![Page 28: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/28.jpg)
Results after Normalization
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 18 / 19
![Page 29: The E ect on Citation Inequality of Di erences in Citation ... · The E ect on Citation Inequality of Di erences in Citation Practices across Scienti c Fields Juan A. Crespo 1, unrongY](https://reader035.vdocument.in/reader035/viewer/2022071014/5fcd170e3b88ee09e246f88a/html5/thumbnails/29.jpg)
Thank you!
Juan A. Crespo, Yunrong Li, Javier Ruiz-Castillo (UC3M)Di�erences in Citation Impacts October 10, 2013 19 / 19