1. No category

Quantifying patterns
of scientific
excellence
Roberta Sinatra
Percy Spencer
Percy Spencer
“What’s the opposite
of ‘Eureka’?”
are there patterns
of excellence in
?
evolution of perfoRmance
in psychology
Ericsson, et al. Psychological review 100 , 363 (1993).
Ericsson et al., Harvard Business Review 85, 114 (2007)
Jones & Weinberg. PNAS, 108(47), 18910-18914 (2011).
Simonton, Psychological Review 104, 66 (1997).
Nature vs nurture
Simonton, D. F. (1997). Psychological Review 104, 66.
Nature vs nurture
Jones, B. F., & Weinberg, B. A. (2011). PNAS, 108(47), 18910-18914.
Nature vs nurture
Ericsson, K. A. & al. (2007). Harvard Business Review 85, 114 .
Ericsson, K. A. & al. (1993). Psychological review 100, 363.
Gladwell, M. (2008). Outliers: The story of success. Penguin UK.
?
We study scientific success
through publications and citations
All papers
All
citations
Kenneth G. Wilson
400
Nobel in physics in 1982
c (citations)
300
200
100
0
0
10
20
30
time (in years)
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
Kenneth G. Wilson
Nobel in physics in 1982
400
c
c (citations)
300
⇤
200
100
0
0
t
⇤
10
20
30
time (in years)
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
timing of the hit is high between 0
and 20 years, decays afterwards
0.2
Data
Randomized
P (t⇤ )
0.15
0.1
0.05
0
0
10
20
t
⇤
30
40
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
are there patterns before
and after the scientific hits?
there are patterns of Productivity
4
hn (t)i
3
2
1
−10 −5
t*
5
time in years
10
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
there are patterns of Productivity
4
hn (t)i
3
High Impact
(c∗10 ≥ 200 )
Middle Impact (20 < c∗10 < 200 )
(c∗10 ≤ 20 )
Low Impact
2
1
−10 −5
t*
5
time in years
10
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
there are patterns of Productivity
4
hn (t)i
3
High Impact
(c∗10 ≥ 200 )
Middle Impact (20 < c∗10 < 200 )
(c∗10 ≤ 20 )
Low Impact
2
1
−10 −5
t*
5
time in years
10
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
there are patterns of Productivity
4
hn (t)i
3
High Impact
(c∗10 ≥ 200 )
Middle Impact (20 < c∗10 < 200 )
(c∗10 ≤ 20 )
Low Impact
2
1
−10 −5
t*
5
time in years
10
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
there are patterns of Productivity
4
hn (t)i
3
High Impact
(c∗10 ≥ 200 )
Middle Impact (20 < c∗10 < 200 )
(c∗10 ≤ 20 )
Low Impact
2
1
−10 −5
t*
5
time in years
10
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
there are patterns of Productivity
4
hn (t)i
3
High Impact
(c∗10 ≥ 200 )
Middle Impact (20 < c∗10 < 200 )
(c∗10 ≤ 20 )
Low Impact
2
1
−10 −5
t*
5
time in years
10
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
there are no patterns of impact
300
hcitationsi
250
200
150
100
50
0
High Impact
−8 −6 −4 −2
N*
2
4
6
8
(c∗10 ≥ 200 )
Middle Impact (20 < c∗10 < 200 )
(c∗10 ≤ 20 )
Low Impact
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
there are no patterns of impact
300
hcitationsi
250
200
150
100
50
0
High Impact
−8 −6 −4 −2
N*
2
4
6
8
(c∗10 ≥ 200 )
Middle Impact (20 < c∗10 < 200 )
(c∗10 ≤ 20 )
Low Impact
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
there are no patterns of impact
300
hcitationsi
250
200
150
100
50
0
High Impact
−8 −6 −4 −2
N*
2
4
6
8
(c∗10 ≥ 200 )
Middle Impact (20 < c∗10 < 200 )
(c∗10 ≤ 20 )
Low Impact
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
there are no patterns of impact
300
hcitationsi
250
200
150
100
50
0
High Impact
−8 −6 −4 −2
N*
2
4
6
8
(c∗10 ≥ 200 )
Middle Impact (20 < c∗10 < 200 )
(c∗10 ≤ 20 )
Low Impact
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
We can’t see the hit coming.
Nor do we learn from it
Timing of the hit is random
0.2
Data
Randomized
P (t⇤ )
0.15
0.1
0.05
0
0
10
20
t
⇤
30
40
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
Timing of the hit is random
0.2
Data
Randomized
P (t⇤ )
0.15
0.1
0.05
0
0
10
20
t
⇤
30
40
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
Impact is random within
a scientific career
Frank G. Wilczek
Physics Nobel, 2004
John B. Fenn
Chemistry Nobel, 2002
6
5
4
3
Mean
2
10
1
10
Productivity
2
N
∗
highest impact
log cpaper
log c⇤
highest impact paper log c⇤
there are statistical Regularities
comparing scientists
8
6
4
2
0
0
Mean
1
2
3
−∗
!log
c
average impact"hlog c
4
⇤
i
average expected maximum grows
with the number of trials
1 toss
100 toss
hs
max
hs
i = 3.5
max
i⇠6
6
5
4
3
Mean
Null Model
2
10
1
Productivity
10
2
N
highest impact paper log c⇤
highest impact paper log c⇤
there are statistical Regularities
comparing scientists
8
6
4
2
0
0
Mean
Null Model
1
2
3
average impact hlog c
⇤
4
i
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
Individuals have different
impact distributions
1
one scientist
P(
c)
0.8
0.6
0.4
N=100
0.2
A third
scientist
0 0
10
1
10
2
10
c
Another Scientist
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
Individuals draw paper impact
from different boxes
?
?
?
a universal stochastic
principle driving
individual impact ?
Excellence model
cij = pj Ei
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
Excellence model
cij = pj Ei
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
Excellence model
cij = pj Ei
3d lognormal
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
Excellence model
cij = pj Ei
3d lognormal
P (p)P (E, N )
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
Excellence model
cij = pj Ei
P (p)
uncoupled
P (E)
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
Excellence model
cij = pj Ei
Universal
P (p)
uncoupled
P (E)
Everyone has the same initial chance to
make a major discovery
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
Untangling skill and luck
P(p)
Data
P(
c)
1
P(p)
0.8
0.6
0.4
0.2
0
10
0
10
1
10
2
is universal
Untangling skill and luck
P(p)
is universal
Data
1
P(p)
c/E)
0.8
0.6
0.4
P(
P(
c)
1
Rescaled Data
0.2
0
P(p)
0.8
0.6
0.4
0.2
10
0
10
1
10
2
0
10
0
10
1
10
2
Excellence model
a
Increasing E
1
c10
13.9
8.1
0.6
4.8
c10
10,i
0.4
2.8
0.2
0
b
1.6
c10
)
PP (≥( c c)
0.8
E
1
10
500
c
c10,i
c
d
300
250
200
150
100
50
0
300
250
200
150
100
50
0
300
250
200
150
100
50
0
E=8.48
1
10
20
30
E=3.49
1
10
20
30
E=1.34
1
10
20
30
time (in years)
8
6
6
5
log c∗
log c∗
The excellence model works
4
2
0
0
1
2
!log c
Mean
Excellence
−∗
3
"
4
3
4
Mean
Excellence
2
10
1
N
10
2
Sustained excellence
detects Nobel Laureates
Nobel Laureates
1
0.8
0.6
0.4
excellence E, 0.98
total citations Ctot, 0.94
h−index, 0.93
highest impact c∗10, 0.92
productivity N, 0.71
0.2
0
0
0.2
0.4
0.6
0.8
1
Other scientists
Sinatra, Wang, Deville, Song, Barabási, in review (2014)
It is not (just) a new indicator,
rather it explains all other
indicators
1.
Impact is random within a
scientist’s career
2.
Sustained excellence:
a universal principle for
individual impact
Team
László Barabási
Pierre Deville
Chaoming Song
Dashun Wang
Thank you!
@robysinatra
www.robertasinatra.com