5 June 2008 | www.nature.com/nature | $10 THE INTERNATIONAL WEEKLY JOURNAL OF SCIENCE NATUREJOBS Alabama bound ● Laws for human motion ● Lévy flight for foraging PATTERNS OF MOBILITY PHOENIX FALLING A postcard from Mars SATURN’S SATELLITES Mayhem in the F ring OBESITY Fat cell numbers are for life
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
453, 697–8
22 5 June 200
8w
ww
.nature.com/nature
no.7196
5 June 2008 | www.nature.com/nature | $10
0 71486 03070 6
2 3
$10.00US $12.99CAN
THE INTERNATIONAL WEEKLY JOURNAL OF SCIENCE
NATUREJOBSAlabama bound
● Laws for human motion● Lévy flight for foraging
PATTERNS OF MOBILITY
PHOENIX FALLING A postcard from Mars
SATURN’S SATELLITESMayhem in the F ring
OBESITYFat cell numbers are for life
������������ ���������� ��������������������
LETTERS
Understanding individual human mobility patternsMarta C. Gonzalez1, Cesar A. Hidalgo1,2 & Albert-Laszlo Barabasi1,2,3
Despite their importance for urban planning1, traffic forecasting2
and the spread of biological3–5 and mobile viruses6, our under-standing of the basic laws governing human motion remainslimited owing to the lack of tools to monitor the time-resolvedlocation of individuals. Here we study the trajectory of 100,000anonymized mobile phone users whose position is tracked for asix-month period. We find that, in contrast with the random tra-jectories predicted by the prevailing Levy flight and random walkmodels7, human trajectories show a high degree of temporal andspatial regularity, each individual being characterized by a time-independent characteristic travel distance and a significant prob-ability to return to a few highly frequented locations. Aftercorrecting for differences in travel distances and the inherentanisotropy of each trajectory, the individual travel patterns col-lapse into a single spatial probability distribution, indicating that,despite the diversity of their travel history, humans follow simplereproducible patterns. This inherent similarity in travel patternscould impact all phenomena driven by human mobility, fromepidemic prevention to emergency response, urban planningand agent-based modelling.
Given the many unknown factors that influence a population’smobility patterns, ranging from means of transportation to job- andfamily-imposed restrictions and priorities, human trajectories are oftenapproximated with various random walk or diffusion models7,8.Indeed, early measurements on albatrosses9, followed by more recentdata on monkeys and marine predators10,11, suggested that animal tra-jectory is approximated by a Levy flight12,13—a random walk for whichstep sizeDr follows a power-law distribution P(Dr) ,Dr2(1 1 b), wherethe displacement exponent b , 2. Although the Levy statistics for someanimals require further study14, this finding has been generalized tohumans7, documenting that the distribution of distances between con-secutive sightings of nearly half-a-million bank notes is fat-tailed. Giventhat money is carried by individuals, bank note dispersal is a proxyfor human movement, suggesting that human trajectories are bestmodelled as a continuous-time random walk with fat-tailed displace-ments and waiting-time distributions7. A particle following a Levyflight has a significant probability to travel very long distances in asingle step12,13, which seems to be consistent with human travel pat-terns: most of the time we travel only over short distances, betweenhome and work, whereas occasionally we take longer trips.
Each consecutive sighting of a bank note reflects the compositemotion of two or more individuals who owned the bill betweentwo reported sightings. Thus, it is not clear whether the observed dis-tribution reflects the motion of individual users or some previouslyunknown convolution between population-based heterogeneities andindividual human trajectories. Contrary to bank notes, mobile phonesare carried by the same individual during his/her daily routine, offeringthe best proxy to capture individual human trajectories15–19.
We used two data sets to explore the mobility pattern of indivi-duals. The first (D1) consisted of the mobility patterns recorded over
a six-month period for 100,000 individuals selected randomly from asample of more than 6 million anonymized mobile phone users. Eachtime a user initiated or received a call or a text message, the location ofthe tower routeing the communication was recorded, allowing usto reconstruct the user’s time-resolved trajectory (Fig. 1a, b). Thetime between consecutive calls followed a ‘bursty’ pattern20 (seeSupplementary Fig. 1), indicating that although most consecutivecalls are placed soon after a previous call, occasionally there are longperiods without any call activity. To make sure that the obtainedresults were not affected by the irregular call pattern, we also studieda data set (D2) that captured the location of 206 mobile phone users,recorded every two hours for an entire week. In both data sets, thespatial resolution was determined by the local density of the morethan 104 mobile towers, registering movement only when the usermoved between areas serviced by different towers. The average ser-vice area of each tower was approximately 3 km2, and over 30% of thetowers covered an area of 1 km2 or less.
To explore the statistical properties of the population’s mobilitypatterns, we measured the distance between user’s positions at con-secutive calls, capturing 16,264,308 displacements for the D1 and10,407 displacements for the D2 data set. We found that the distri-bution of displacements over all users is well approximated by atruncated power-law:
P Drð Þ~ DrzDr0ð Þ{bexp {Dr=kð Þ ð1Þ
with exponent b 5 1.75 6 0.15 (mean 6 standard deviation),Dr0 5 1.5 km and cutoff values k D1
j ~400 km and k D2j ~80 km
(Fig. 1c, see the Supplementary Information for statistical valida-tion). Note that the observed scaling exponent is not far fromb 5 1.59 observed in ref. 7 for bank note dispersal, suggesting thatthe two distributions may capture the same fundamental mechanismdriving human mobility patterns.
Equation (1) suggests that human motion follows a truncatedLevy flight7. However, the observed shape of P(Dr) could be explainedby three distinct hypotheses: first, each individual follows a Levy tra-jectory with jump size distribution given by equation (1) (hypothesisA); second, the observed distribution captures a population-basedheterogeneity, corresponding to the inherent differences between indi-viduals (hypothesis B); and third, a population-based heterogeneitycoexists with individual Levy trajectories (hypothesis C); hence, equa-tion (1) represents a convolution of hypotheses A and B.
To distinguish between hypotheses A, B and C, we calculated theradius of gyration for each user (see Supplementary Information), inter-preted as the characteristic distance travelled by user a when observed upto time t (Fig. 1b). Next, we determined the radius of gyration distri-bution P(rg) by calculating rg for all users in samples D1 and D2, findingthat they also can be approximated with a truncated power-law:
P rg
� �~ rgzr0
g
� �{br
exp {rg
�k
� �ð2Þ
1Center for Complex Network Research and Department of Physics, Biology and Computer Science, Northeastern University, Boston, Massachusetts 02115, USA. 2Center for ComplexNetwork Research and Department of Physics and Computer Science, University of Notre Dame, Notre Dame, Indiana 46556, USA. 3Center for Cancer Systems Biology, Dana FarberCancer Institute, Boston, Massachusetts 02115, USA.
with r0g ~5:8 km, br 5 1.65 6 0.15 and k 5 350 km (Fig. 1d, see
Supplementary Information for statistical validation). Levy flightsare characterized by a high degree of intrinsic heterogeneity, raisingthe possibility that equation (2) could emerge from an ensemble ofidentical agents, each following a Levy trajectory. Therefore, wedetermined P(rg) for an ensemble of agents following a random walk(RW), Levy flight (LF) or truncated Levy flight (TLF) (Fig. 1d)8,12,13.We found that an ensemble of Levy agents display a significant degreeof heterogeneity in rg; however, this was not sufficient to explain thetruncated power-law distribution P(rg) exhibited by the mobilephone users. Taken together, Fig. 1c and d suggest that the differencein the range of typical mobility patterns of individuals (rg) has astrong impact on the truncated Levy behaviour seen in equation(1), ruling out hypothesis A.
If individual trajectories are described by an LF or TLF, thenthe radius of gyration should increase with time as rg(t) , t3/(2 1 b)
(ref. 21), whereas, for an RW, rg(t) , t1/2; that is, the longer weobserve a user, the higher the chance that she/he will travel to areasnot visited before. To check the validity of these predictions, wemeasured the time dependence of the radius of gyration for userswhose gyration radius would be considered small (rg(T) # 3 km),medium (20 , rg(T) # 30 km) or large (rg(T) . 100 km) at the endof our observation period (T 5 6 months). The results indicate that
the time dependence of the average radius of gyration of mobilephone users is better approximated by a logarithmic increase, notonly a manifestly slower dependence than the one predicted by apower law but also one that may appear similar to a saturationprocess (Fig. 2a and Supplementary Fig. 4).
In Fig. 2b, we chose users with similar asymptotic rg(T) afterT 5 6 months, and measured the jump size distribution P(Drjrg)for each group. As the inset of Fig. 2b shows, users with small rg travelmostly over small distances, whereas those with large rg tend todisplay a combination of many small and a few larger jump sizes.Once we rescaled the distributions with rg (Fig. 2b), we found that thedata collapsed into a single curve, suggesting that a single jump sizedistribution characterizes all users, independent of their rg. Thisindicates that P Dr rg
��� �*r{a
g F Dr�
rg
� �, where a < 1.2 6 0.1 and
F(x) is an rg-independent function with asymptotic behaviour, thatis, F(x) , x2a for x , 1 and F(x) rapidly decreases for x? 1.Therefore, the travel patterns of individual users may be approxi-mated by a Levy flight up to a distance characterized by rg. Mostimportant, however, is the fact that the individual trajectories arebounded beyond rg; thus, large displacements, which are the sourceof the distinct and anomalous nature of Levy flights, are statisticallyabsent. To understand the relationship between the different expo-nents, we note that the measured probability distributions are related
Figure 1 | Basic human mobility patterns. a, Week-long trajectory of 40mobile phone users indicates that most individuals travel only over shortdistances, but a few regularly move over hundreds of kilometres. b, Thedetailed trajectory of a single user. The different phone towers are shown asgreen dots, and the Voronoi lattice in grey marks the approximate receptionarea of each tower. The data set studied by us records only the identity of theclosest tower to a mobile user; thus, we can not identify the position of a userwithin a Voronoi cell. The trajectory of the user shown in b is constructedfrom 186 two-hourly reports, during which the user visited a total of 12different locations (tower vicinities). Among these, the user is found on 96and 67 occasions in the two most preferred locations; the frequency of visits
for each location is shown as a vertical bar. The circle represents the radius ofgyration centred in the trajectory’s centre of mass. c, Probability densityfunction P(Dr) of travel distances obtained for the two studied data sets D1
and D2. The solid line indicates a truncated power law for which theparameters are provided in the text (see equation (1)). d, The distributionP(rg) of the radius of gyration measured for the users, where rg(T) wasmeasured after T 5 6 months of observation. The solid line represents asimilar truncated power-law fit (see equation (2)). The dotted, dashed anddot-dashed curves show P(rg) obtained from the standard null models (RW,LF and TLF, respectively), where for the TLF we used the same step sizedistribution as the one measured for the mobile phone users.
tary Information) that up to the leading order we haveb 5 br 1 a 2 1, consistent, within error bars, with the measuredexponents. This indicates that the observed jump size distributionP(Dr) is in fact the convolution between the statistics of individualtrajectories P(Drgjrg) and the population heterogeneity P(rg), con-sistent with hypothesis C.
To uncover the mechanism stabilizing rg, we measured thereturn probability for each individual Fpt(t) (first passage timeprobability)21,22, defined as the probability that a user returns to theposition where he/she was first observed after t hours (Fig. 2c). For atwo-dimensional random walk, Fpt(t) should follow ,1/(t ln2(t))(ref. 21). In contrast, we found that the return probability is char-acterized by several peaks at 24 h, 48 h and 72 h, capturing a strongtendency of humans to return to locations they visited before,describing the recurrence and temporal periodicity inherent tohuman mobility23,24.
To explore if individuals return to the same location over and over,we ranked each location on the basis of the number of times anindividual was recorded in its vicinity, such that a location withL 5 3 represents the third-most-visited location for the selected indi-vidual. We find that the probability of finding a user at a location witha given rank L is well approximated by P(L) , 1/L, independent of thenumber of locations visited by the user (Fig. 2d). Therefore, peopledevote most of their time to a few locations, although spending theirremaining time in 5 to 50 places, visited with diminished regularity.Therefore, the observed logarithmic saturation of rg(t) is rooted inthe high degree of regularity in the daily travel patterns of individuals,captured by the high return probabilities (Fig. 2b) to a few highlyfrequented locations (Fig. 2d).
An important quantity for modelling human mobility patterns isthe probability density function Wa(x, y) to find an individual a in agiven position (x, y). As it is evident from Fig. 1b, individuals live andtravel in different regions, yet each user can be assigned to a welldefined area, defined by home and workplace, where she or he canbe found most of the time. We can compare the trajectories of dif-ferent users by diagonalizing each trajectory’s inertia tensor, provid-ing the probability of finding a user in a given position (see Fig. 3a) inthe user’s intrinsic reference frame (see Supplementary Informationfor the details). A striking feature of W (x, y) is its prominent spatialanisotropy in this intrinsic reference frame (note the different scalesin Fig. 3a); we find that the larger an individual’s rg, the more pro-nounced is this anisotropy. To quantify this effect, we defined theanisotropy ratio S ; sy/sx, where sx and sy represent the standarddeviation of the trajectory measured in the user’s intrinsic referenceframe (see Supplementary Information). We found that S decreasesmonotonically with rg (Fig. 3c), being well approximated withS*r{g
g for g < 0.12. Given the small value of the scaling exponent,other functional forms may offer an equally good fit; thus, mecha-nistic models are required to identify if this represents a true scalinglaw or only a reasonable approximation to the data.
To compare the trajectories of different users, we removed theindividual anisotropies, rescaling each user trajectory with itsrespective sx and sy. The rescaled ~WW x=sx ,y
�sy
� �distribution
(Fig. 3b) is similar for groups of users with considerably differentrg, that is, after the anisotropy and the rg dependence are removedall individuals seem to follow the same universal ~WW ~xx,~yyð Þ probabi-lity distribution. This is particularly evident in Fig. 3d, where weshow the cross section of ~WW x=sx ,0ð Þ for the three groups ofusers, finding that apart from the noise in the data the curves areindistinguishable.
Taken together, our results suggest that the Levy statistics observedin bank note measurements capture a convolution of the populationheterogeneity shown in equation (2) and the motion of individualusers. Individuals display significant regularity, because they returnto a few highly frequented locations, such as home or work. Thisregularity does not apply to the bank notes: a bill always followsthe trajectory of its current owner; that is, dollar bills diffuse, buthumans do not.
The fact that individual trajectories are characterized by thesame rg-independent two-dimensional probability distribution~WW x=sx ,y
�sy
� �suggests that key statistical characteristics of indi-
vidual trajectories are largely indistinguishable after rescaling.Therefore, our results establish the basic ingredients of realisticagent-based models, requiring us to place users in number propor-tional with the population density of a given region and assign eachuser an rg taken from the observed P(rg) distribution. Using thepredicted anisotropic rescaling, combined with the density function~WW x,yð Þ, the shape of which is provided as Supplementary Table 1,we can obtain the likelihood of finding a user in any location. Giventhe known correlations between spatial proximity and social links,our results could help quantify the role of space in network develop-ment and evolution25–29 and improve our understanding of diffusionprocesses8,30.
1 3 4 5 6
0.2
0.4
0.6
1 10 1000.001
0.01
0.1
1.0
5 loc.10 loc.30 loc.50 loc.~(L)–1
L
L
P(L
)
P(L
) 2
c d
ba
<r
g(t)>
(km
)
1
10
100
0 1,000 2,000 3,000 4,000
t (h)
RWLF
rg ≤ 3 km20 < rg ≤ 30 km
rg > 100 km
UsersModels
10–3 10–2 10–1 100 101 10210–5
10–4
10–3
10–2
10–1
100
101
102 rg~4rg~10rg~40rg~100rg~200
100 101 102 103 10410–610–510–410–310–210–1100
P (∆
r|rg)
∆r (km)
∆r/rg
r g P
(∆r|r
g)
F pt(t
)
0 24 48 72 96 120 144 168 192 216 2400
0.01
0.02
t (h)
UsersRW
α
Slope –1.2
Figure 2 | The bounded nature of human trajectories. a, Radius of gyrationÆrg(t)æ versus time for mobile phone users separated into three groupsaccording to their final rg(T), where T 5 6 months. The black curvescorrespond to the analytical predictions for the random walk models,increasing with time as Ærg(t)æ | LF,TLF , t3/2 1 b (solid curve) andÆrg(t)æ | RW , t0.5 (dotted curve). The dashed curves corresponding to alogarithmic fit of the form A 1 B ln(t), where A and B are time-independentcoefficients that depend on rg. b, Probability density function of individualtravel distances P(Dr | rg) for users with rg 5 4, 10, 40, 100 and 200 km. As theinset shows, each group displays a quite different P(Dr | rg) distribution. Afterrescaling the distance and the distribution with rg (main panel), the differentcurves collapse. The solid line (power law) is shown as a guide to the eye.c, Return probability distribution, Fpt(t). The prominent peaks capture thetendency of humans to return regularly to the locations they visited before,in contrast with the smooth asymptotic behaviour ,1/(t ln(t)2) (solid line)predicted for random walks. d, A Zipf plot showing the frequency of visitingdifferent locations (loc.). The symbols correspond to users that have beenobserved to visit nL 5 5, 10, 30 and 50 different locations. Denoting with (L),the rank of the location listed in the order of the visit frequency, the data arewell approximated by R(L) , L21. The inset is the same plot in linear scale,illustrating that 40% of the time individuals are found at their first twopreferred locations; bars indicate the standard error.
Received 19 December 2007; accepted 27 March 2008.
1. Horner, M. W. & O’Kelly, M. E. S Embedding economies of scale concepts for hubnetworks design. J. Transp. Geogr. 9, 255–265 (2001).
2. Kitamura, R., Chen, C., Pendyala, R. M. & Narayaran, R. Micro-simulation of dailyactivity-travelpatternsfortraveldemandforecasting.Transportation 27,25–51(2000).
3. Colizza, V., Barrat, A., Barthelemy, M., Valleron, A.-J. & Vespignani, A. Modelingthe worldwide spread of pandemic influenza: baseline case and containmentinterventions. PLoS Medicine 4, 95–110 (2007).
4. Eubank, S. et al. Controlling epidemics in realistic urban social networks. Nature429, 180–184 (2004).
5. Hufnagel, L., Brockmann, D. & Geisel, T. Forecast and control of epidemics in aglobalized world. Proc. Natl Acad. Sci. USA 101, 15124–15129 (2004).
6. Kleinberg, J. The wireless epidemic. Nature 449, 287–288 (2007).7. Brockmann, D. D., Hufnagel, L. & Geisel, T. The scaling laws of human travel.
Nature 439, 462–465 (2006).8. Havlin, S. & Ben-Avraham, D. Diffusion in disordered media. Adv. Phys. 51,
187–292 (2002).9. Viswanathan, G. M. et al. Levy flight search patterns of wandering albatrosses.
Nature 381, 413–415 (1996).10. Ramos-Fernandez, G. et al. Levy walk patterns in the foraging movements of
spider monkeys (Ateles geoffroyi). Behav. Ecol. Sociobiol. 273, 1743–1750 (2004).11. Sims, D. W. et al. Scaling laws of marine predator search behaviour. Nature 451,
1098–1102 (2008).12. Klafter, J., Shlesinger, M. F. & Zumofen, G. Beyond brownian motion. Phys. Today
49, 33–39 (1996).13. Mantegna, R. N. & Stanley, H. E. Stochastic process with ultraslow convergence to
a gaussian: the truncated Levy flight. Phys. Rev. Lett. 73, 2946–2949 (1994).14. Edwards, A. M. et al. Revisiting Levy flight search patterns of wandering
albatrosses, bumblebees and deer. Nature 449, 1044–1049 (2007).15. Sohn, T. et al. in Proc. 8th Int. Conf. UbiComp 2006 212–224 (Springer, Berlin, 2006).16. Onnela, J.-P. et al. Structure and tie strengths in mobile communication networks.
Proc. Natl Acad. Sci. USA 104, 7332–7336 (2007).17. Gonzalez, M. C. & Barabasi, A.-L. Complex networks: from data to models. Nature
Physics 3, 224–225 (2007).18. Palla, G., Barabasi, A.-L. & Vicsek, T. Quantifying social group evolution. Nature
446, 664–667 (2007).19. Hidalgo, C. A. & Rodriguez-Sickert, C. The dynamics of a mobile phone network.
Physica A 387, 3017–3024 (2008).
20. Barabasi, A.-L. The origin of bursts and heavy tails in human dynamics. Nature435, 207–211 (2005).
21. Redner, S. A Guide to First-Passage Processes (Cambridge Univ. Press, Cambridge,UK, 2001).
22. Condamin, S., Benichou, O., Tejedor, V. & Klafter, J. First-passage times incomplex scale-invariant media. Nature 450, 77–80 (2007).
23. Schlich, R. & Axhausen, K. W. Habitual travel behaviour: evidence from a six-weektravel diary. Transportation 30, 13–36 (2003).
24. Eagle, N. & Pentland, A. Eigenbehaviours: identifying structure in routine. Behav.Ecol. Sociobiol. (in the press).
25. Yook, S.-H., Jeong, H. & Barabasi, A. L. Modeling the Internet’s large-scaletopology. Proc. Natl Acad. Sci. USA 99, 13382–13386 (2002).
26. Caldarelli, G. Scale-Free Networks: Complex Webs in Nature and Technology.(Oxford Univ. Press, New York, 2007).
27. Dorogovtsev, S. N. & Mendes, J. F. F. Evolution of Networks: From Biological Nets tothe Internet and WWW (Oxford Univ. Press, New York, 2003).
28. Song, C. M., Havlin, S. & Makse, H. A. Self-similarity of complex networks. Nature433, 392–395 (2005).
29. Gonzalez, M. C., Lind, P. G. & Herrmann, H. J. A system of mobile agents to modelsocial networks. Phys. Rev. Lett. 96, 088702 (2006).
30. Cecconi, F., Marsili, M., Banavar, J. R. & Maritan, A. Diffusion, peer pressure, andtailed distributions. Phys. Rev. Lett. 89, 088102 (2002).
Supplementary Information is linked to the online version of the paper atwww.nature.com/nature.
Acknowledgements We thank D. Brockmann, T. Geisel, J. Park, S. Redner,Z. Toroczkai, A. Vespignani and P. Wang for discussions and comments on themanuscript. This work was supported by the James S. McDonnell Foundation 21stCentury Initiative in Studying Complex Systems, the National Science Foundationwithin the DDDAS (CNS-0540348), ITR (DMR-0426737) and IIS-0513650programs, and the US Office of Naval Research Award N00014-07-C. Dataanalysis was performed on the Notre Dame Biocomplexity Cluster supported inpart by the NSF MRI grant number DBI-0420980. C.A.H. acknowledges supportfrom the Kellogg Institute at Notre Dame.
Author Information Reprints and permissions information is available atwww.nature.com/reprints. Correspondence and requests for materials should beaddressed to A.-L.B. ([email protected]).
–15 0 15–15
0
15
–150 0 150–150
0
150
–1,200 0 1,200–1,200
0
1,200
Fmin
Fmax
x (km) x (km) x (km)y
(km
)
y (k
m)
y (k
m)
y/s y
y/s y
x/sx
0.1
0.2
0.3
0.4
s y/s
x
10 100 1,0001rg (km)
–15 0 15–15
0
15
–15
0
15
–15
0
15
–15 0 15 –15 0 15
–10 –5 0 5 1010–6
10–4
10–2
100
(x/s
x,0)
rg ≤ 3 km
20 km < rg < 30 km
rg > 100 km
~
10–2
10–3
10–4
10–5
10–6
a
b
dc
y/s y
x/sx x/sx
x/sx
F
Figure 3 | The shape of human trajectories.a, The probability density function W(x, y) offinding a mobile phone user in a location (x, y) inthe user’s intrinsic reference frame (seeSupplementary Information for details). Thethree plots, from left to right, were generated for10,000 users with: rg # 3, 20 , rg # 30 andrg . 100 km. The trajectories become moreanisotropic as rg increases. b, After scaling eachposition with sx and sy, the resulting~WW x=sx ,y
�sy
� �has approximately the same shape
for each group. c, The change in the shape ofW(x, y) can be quantified calculating the isotropyratio S ; sy/sx as a function of rg, which decreasesas S*r{0:12
g (solid line). Error bars represent thestandard error. d, ~WW x=sx ,0ð Þ representing thex-axis cross-section of the rescaled distribution~WW x=sx ,y
S.V., da Luz, M.G.E., Raposo, E. P., Stanley, H. E., & Viswanathan, G. M. Revisiting Lvy flight search
patterns of wandering albatrosses, bumblebees and deer.Nature449, 1044-1049 (2007).
[8] Sims D.W., Righton D. & Pitchford, J.W. Minimizing errors in identifying Levy flight behaviour of
organisms.Journal of Animal Ecology76, 222-229 (2007).
[9] Pueyo, S. & Jovani, R. Comment on A keystone mutualism drives pattern in a power function.Science
313, 1739c (2006).
[10] Clauset, A., Rohila Shalizi, C. & Newman, M.E.J. Power-law distributions in empirical data.
arXiv:physics:/07061062.
[11] Brockmann, D., Hufnagel, L. & Geisel, T. The scaling laws of human travel.Nature439, 462-465
only mean that morale at places such as the Environmental Protec-tion Agency will improve.
So far, McCain and Obama have set out relatively different plat-forms on science and technology issues. For his part, Obama has yet to give a substantive speech on science issues, as Clinton did on the 50th anniversary of Sputnik. But he has adopted many of the traditional Democratic platforms, such as increasing federal funding for bio-medical research and improving the jobs pipeline for young scientists. He has put a strong emphasis on the importance of technology for improving the lives of everyday Americans and their access to government. And he has been remarkably successful at using the Internet to turn his supporters into active participants in the political process. An Obama administra-tion might well mean that more technologically savvy people will be drawn into public life — and perhaps that more young people will choose science and engineering careers.
In contrast, McCain has revealed few details of his science and
technology agenda. But in some areas he has been quite outspoken. Last week, for example, he called for a new treaty to reduce the US and Russian nuclear arsenals, suggesting he would lower the number of weapons even beyond the cuts planned by the Bush administration.
And on 12 May he outlined a detailed plan for con-trolling greenhouse-gas emissions — the centrepiece of which is a cap-and-trade system of the type he has been advocating since at least 2003, when he co-spon-sored the first meaningful bill on that subject.
In the end, the main factor is not how Obama or McCain feels about specific science-related issues.
What American voters deserve to know is how each candidate’s mind works. Does he listen to a handful of ideology-driven advisers in making key decisions? Or does he look facts in the face and base his conclusions on all the available evidence? If it’s the latter, then that is the candidate who is in sync with science at a level far more meaningful than any immediate argument over research budgets or competitiveness. ■
A flood of hard dataSocial scientists have a new handle on group behaviour — but its causes remain a challenge.
Every human being makes choices, takes action and is affected by the environment in a way that seems utterly idiosyncratic. Yet in the aggregate, as the British philosopher John Stuart
Mill put it a century and a half ago, human events “most capricious and uncertain” can take on “a degree of regularity approaching to mathematical”. A case in point is the analysis of mobile-phone data discussed by González et al. on page 779 of this issue. It reveals just such a mathematical regularity in the seemingly unpredictable way people move around during their daily lives (see also page 714).
As remarkable as this result is — and it is still not completely understood — the research is just as notable for its methodology. Social scientists have long struggled with a paucity of hard data about human activities; people’s self-reporting about their social interactions, say, or their move-ment patterns is labour-intensive to collect and notoriously unreliable. In this case, the researchers obtained objective data on individuals’ movements from mobile-phone networks (albeit without access to any individual’s identity, for privacy reasons). This gave them a data set of proportions almost unheard of for such a complex aspect of behaviour: more than 16 million ‘hops’ for 100,000 people. The resulting statistics show a strikingly small scatter, giving grounds for confidence in the mathematical laws they disclose.
The mobile-phone technique is simply the latest example of how modern information technologies are giving social scientists the power to make measurements that are often as precise as those in the ‘hard’ sciences. By analysing e-mail transmissions, for example, or doing automated searches of publication databases,
social scientists can collect detailed information on the network structure of scientific collaborations and other social interactions. And by allowing their subjects to interact online, researchers can do large-scale studies of, for example, the role of social inter-actions in opinion formation, complete with control groups and tuneable parameters.
It’s not an overstatement to say that these tools are fostering a whole new type of social science — with applications that go well beyond the conventional boundaries of the field. There is sure to be commercial interest in the detailed patterns of usage for portable electronics, for example, and the nature of mass human movement could inform urban planning and the development of transporta-tion networks. Epidemiologists, meanwhile, will no longer be forced to work with highly oversimplified models of infection rates and disease spread: recent work has clarified how the transmission of
disease depends on the precise structural details of the network of person-to-person contacts.
For all their promise, making sense of these new data sets requires a rather different set of statisti-cal skills than those needed in conventional social science — which may be one reason why studies such as that by González et al. are so often con-ducted by researchers trained in the physical sci-ences. To some extent this ‘physicalization’ of the social sciences is healthy for the field; it has already
brought in many new ideas and perspectives. But it also needs to be regarded with some caution.
As many social scientists have pointed out, the goal of their disci-pline is not simply to understand how people behave in large groups, but to understand what motivates individuals to behave the way they do. The field cannot lose focus on that — even as it moves to exploit the power of these new technological tools, and the mathematical regularities they reveal. Comprehending capricious and uncertain human events at every level remains one of the most challenging questions in science. ■
“There can be little doubt that the next US administration will be more science-friendly than the current one.”
“The goal of social science is not simply to understand how people behave in large groups, but to understand what motivates individuals to behave the way they do.”
698
NATURE|Vol 453|5 June 2008EDITORIALS
Food is not, in general, spread equally around the world; it comes in lumps. Foragers thus need a strategy for find-ing those lumps. One appealing option
is a Lévy flight — a mathematical concept used in physics. Lévy flights are many-legged jour-neys in which most of the legs are short, but a few are much longer. They are found in some sorts of diffusion, in fluid turbulence, even in astrophysics. In animal behaviour, the longer the flight, the farther afield a creature will get, offering a way to efficiently exploit food nearby but also to discover sources farther away.
“The pattern captures what biologists often
notice,” says behavioural ecologist David Sims of the Marine Biological Association Labora-tory in Plymouth, UK. “Animals often take lots of short steps in a localized area before making long jumps to new areas.”
But just because it makes qualitative sense doesn’t mean it is a mathematical key to the real world. Hard evidence is needed to show that the pattern is a real Lévy flight, in which the frequency of steps of given distances is firmly constrained. And this evidence is what physicist Gandhimohan Viswanathan, then a graduate student at Boston University in Massachusetts, and his colleagues seemed to find in 1996.
Albatrosses soar over tremendous distances as they circle the oceans, alighting here or there to feed on squid, fish or krill before heading off again. Observers had thought the foraging was random; but any hidden pattern would be evident only on the scale of seas and oceans. It was this large pattern that Viswanathan, now at the Federal University of Alagoas in Brazil, decided to look for, using electronic logging data gathered by field ecologists at the British Antarctic Survey (BAS) in Cambridge.
Viswanathan and his colleagues found a scale-free fractal-like pattern in the data1, just what a Lévy flight ought to produce. Three years
C. B
ERRI
ETHE MATHEMATICAL MIRROR
TO ANIMAL NATURE
Animal behaviour is an endless challenge to mathematical modellers. In the first of two features, Mark Buchanan looks at how a mathematical principle from physics
might be able to explain patterns of movement. In the second, Arran Frood asks what current models can teach us about ecological networks half a billion years old.
NATURE|Vol 453|5 June 2008
714
later, they seemed to be on the track of a new principle of ecology when they showed that this way of moving is, under some conditions, theoretically the best way for animals to find scarce prey2. They and other researchers soon reported the same pattern in the movements of everything from reindeer and bumblebees to soil amoebas and the habits of fishermen3. The phenomenon is attracting more and more interest, and it seems to apply to more than just foraging. Research in this week’s Nature shows that it applies to the movements of mobile-phone users4 too (see page 779).
Flights of fancy?There’s just one problem. Although other exam-ples stand up to scrutiny, the one that started the field off does not, at least for now. There’s a lesson in that. When modellers use data from the field, they have to be sure that the data really represent what they think they represent, and that they fit tightly to their model. The devil is in the detail, when sparse data can put almost all conclusions on shaky ground.
The case for Lévy flights by albatrosses ran into problems in 2004, when physicist Sergey Buldyrev, also of Boston Univer-sity and one of Viswanathan’s co-authors on the original albatross paper, analysed new data on albatross movements. The Lévy pattern didn’t turn up. Revisiting the original data collected by the BAS
researchers, Buldyrev discovered that the longest flights recorded, which were crucial to the distinctive fractal fingerprint, might have been artefacts of the recording technique.
The original albatross data came from devices called immersion loggers attached to the birds’ legs. The devices recorded the proportion of time in each quarter-minute that the birds sat on the sea surface. From these data, the research-ers could then infer flights as periods during which the birds remained dry. From five birds, the researchers had obtained a total of 363 flight times, which seemed to show the Lévy pattern.
But Buldyrev wondered whether the long-est periods of dry-leggedness — which always seemed to be the first and last in a bout of movement — might in fact record the birds sitting on their nests. The data had not been saying what the team thought they were saying. Finding that the Lévy pattern vanished when these data points were omitted, Buldyrev and his colleagues wrote up a manuscript and sent a draft to ecologist Richard Phillips at the BAS. Phillips, working with ecological modeller Andrew Edwards, also at the BAS, confirmed that there was no support for Lévy flights. Later, when they discovered that some of the albatrosses also had location trackers fitted to them, the BAS team proved that the birds
weren’t moving during the alleged long flights. “I was disappointed,” says Viswanathan, “but also curious, surprised and perplexed.”
The Lévy flight notion took another blow last October, when the Boston and Cambridge groups collaborated to publish a comprehen-sive reanalysis of the original albatross data, including an analysis of a new data set and a reconsideration of earlier studies of deer and bumblebees5. They found that the deer and bumblebee data were also ambiguous — the deer data, for example, actually reflected time spent cropping and processing food at a par-ticular feeding site, rather than time spent moving between sites. Using improved statisti-cal techniques, the teams found that none of the data offered strong support for the Lévy flight pattern. The results, they say, “question the strength of the empirical evidence for biologi-cal Lévy flights.”
It looked like a simple tale of problematic data corrected. But later last year, Sims and his colleagues presented strong evidence for Lévy-like patterns in the forag-ing of numerous marine predators, including sharks, turtles and penguins6. They used what all researchers agree are more sophisticated statistical methods, and much larger data sets. Sims and others now suggest that the data really do point to Lévy flights for a variety of animals, including humans.
Not everyone yet agrees with this position. But they do agree that the episode illustrates the difficulties inherent in identifying statisti-cal patterns with limited data. The difference between a Lévy flight and a more familiar form of random walk, brownian motion, is the distri-bution of steps of different lengths. In brownian
motion, as seen in the jittering of a pollen grain buffeted on all sides by invisible molecules, the distribution of distances follows a bell-shaped curve, so the size of the next step is at least crudely predictable — it is never 10 or 100 times bigger than the average, for example.
Doing the Lévy walkA Lévy flight is a similar sort of random walk — but the distribution of distances is differ-ent. For example, the probability of large steps of size D might fall off in proportion to dγ, with γ being a number somewhere between
1 and 3. This distribution, in what is known as a power law, gives more frequent long steps than a bell curve, and produces a pattern characterized by lots of smaller movements broken epi-sodically by long excursions.
Diagnosing a true Lévy flight means showing that the power-law distribution holds. There is a simple statistical approach to
this. First ‘bin the data’: that is, count up the events that fall within each small range of dis-tances to get a measure of the way the prob-ability of differing distances is distributed. If a power law holds, the relationship between the logarithm of this probability distribution and the logarithm of the distance will be linear. Hence, if the log of the first is charted against the log of the second, you’ll get a straight line.
As Edwards points out, however, this tech-nique can lead to trouble. “It’s well known that log–log axes tend to make relationships look straight.” The problem is at its worst when data are in short supply. A more rigorous approach, he says, is to decide mathematically which of two possible distributions, say a power law or an exponential, the data fit better. But such
T. A
LLO
FS/Z
EFA
/CO
RBIS
“It was wrong, yet it turned out to be fruitful because it led to other ideas that we now think are correct.” — Eugene Stanley
Dry of foot, but not flying.
715
NATURE|Vol 453|5 June 2008 NEWS FEATURE
determinations need a lot of data.Sims agrees. Inspired by the original alba-
tross paper, he and his colleagues used satellite-linked tags to gather data on plankton-feeding basking sharks. They found horizontal tracks reminiscent of Lévy-like movements, but never obtained enough data to permit a sound statis-tical analysis. “A lack of data,” he says, “means you can fail to detect the pattern even if it’s there, or detect an apparently similar pattern even if it is not.”
Two years ago, Sims hit on the idea to look at sharks’ vertical movements instead. These were recorded at 1-minute intervals for months on end, providing more than 400,000 data points for analysis. Using statistical methods developed in part by Mark Newman of the University of Michigan, Ann Arbor, and similar to those used by Edwards and his team, they found a strong signal of Lévy behaviour. Sims then organized a collaboration of 18 researchers from four coun-tries to gather and test similar data for other marine predators, finding the Lévy pattern for tuna, cod, leatherback turtles and penguins6 .
Sims says that his paper “represents some of the strongest evidence for Lévy-like behaviour in wild predators”. “The debate has shifted,” says Frederic Bartumeus of Princeton University in New Jersey, who in 2003 found Lévy patterns in the movements of plankton. “The question now isn’t whether animals perform Lévy walks, but when they do — and why.”
Although welcoming the use of larger data sets, Edwards, now at the Pacific Biological Sta-tion in Nanaimo, British Columbia, Canada, doesn’t think that these studies end the debate. He says that some scientists have started to use the somewhat softer phrase ‘Lévy-like’ to describe their results, which may make their claims more defensible, but also introduces some vagueness into the discussion. “How ‘non-Lévy-like’ do the data have to be for them not to be considered ‘Lévy-like’ any more?” he says.
The matter is not mere pedantry: getting the pattern right should help researchers to answer meaningful biological questions
— which organisms, if any, forage optimally, and why. Yet for Sims, the qualifier ‘like’ is not without its uses. It could be useful in probing the complex, interacting fac-tors that affect movement patterns. “Animals often undertake other behaviours interspersed with searching, such as social interactions or predator avoidance,” he says,
which may weaken the Lévy signal.
Man in the mirrorHowever the debate plays out, analyses of data from one particular animal, humans, are likely to be increasingly important. Over the past dec-ade, technology has transformed researchers’ ability to gather quantitative data on human activities, ranging from patterns of e-mail use to consumers’ buying habits. People happily carry radio trackers and tags around in the form of mobile phones. “We finally have objective measurements of what people do,” says Albert-László Barabási, a researcher studying human dynamics in this way at the Center for Com-plex Network Research, based at Northeastern University in Boston. “Our observations don’t influence them.”
This work can be viewed, perhaps, as the beginnings of a natural ecol-ogy of human behaviour, for which understanding patterns of physical movement — the crude equivalent of animal for-aging — would offer an obvious first goal. Two years ago, phys-icist Dirk Brockmann of the Max Planck Institute in Göttin-gen, Germany, took an indirect stab at the issue using the website www.wheres-george.com, which facilitates the tracking of dollar notes moving through the United States. People can go to the site and enter the date, their location and the serial numbers of dollar bills in their possession. As the bills move, the site shows their changing locations.
Almost 60% of bills starting in New York City were reported 2 weeks later still within 10 kilo-metres of their starting point. But another 7% had jumped to distances beyond 800 kilome-tres. If this seems similar to the Lévy pattern, it is. The researchers found that the distribution of distances travelled over a short time follows a power law with a γ equal to about 1.6 (ref. 7).
These data don’t directly say anything about the human movements that transport dol-lar bills. But a team led by Barabási has now gone one step further, using anonymized
mobile-phone data to track the movements of more than 100,000 people over a 6-month period. The statistics, they found, again show the Lévy pattern, although with some additional complexity4 .
The team found, overall, that the distribu-tion of the distance moved between two sub-sequent phone calls follows a power law with an exponential cut-off. The best way to explain this pattern, the researchers argue, is through a combination of two effects — first, a real ten-dency for individuals to move in a Lévy-like pattern, with many short movements and less frequent long excursions, but also a difference between people in the overall scale on which they move, with some people being inher-ently longer travellers than others. When the researchers normalized the measurements so that the person-to-person scale factor no longer played a part, the data for all the participants fell onto a single curve. “There are a lot of details that make us different,” says Barabási, “but behind it all there’s a universal pattern.”
And what of the albatrosses? Are they an odd-ity — an error that nevertheless served as the basis for insights into truth? Perhaps. “I think of it like the Bohr model of the atom,” says Eugene Stanley, a physicist from Boston University who was one of the original authors. “It was wrong, yet it turned out to be fruitful. The remarkable fact is that flawed data led to a fascinating idea: a general law governing animal movement.”
Or perhaps albatrosses do roam the high seas in the way that Lévy might have antici-pated, and we will know that in time with better data and anal-yses. As Viswanathan points out, he and his colleagues’ 1999 paper showing the theoretical optimality of Lévy-style forag-ing provides a good a priori
reason to expect that some animals, and quite possibly albatrosses, might exploit this trick.
“Given the power of natural selection,” says Viswanathan, “it seems unlikely to me that Lévy walks wouldn’t exist somewhere in ani-mal biology. It would be as strange as if vision had never evolved.” ■
Mark Buchanan is author of The Social Atom.
1. Viswanathan, G. M. et al. Nature 381, 413–415 (1996).2. Viswanathan, G. M. et al. Nature 401, 911–914 (1999).3. Viswanathan, G. M., Raposob, E. P. & da Luzdoi,
M. G. E. Phys. Life Rev. (in the press) doi:10.1016/j.plrev.2008.03.002
4. González, M. C., Hidalgo, C. A. & Barabási, A.-L. Nature 453, 779–782 (2008).
5. Edwards, A. M. et al. Nature 449, 1044–1048 (2007).6. Sims, D. W. et al. Nature 451, 1098–1102 (2008).7. Brockmann, D., Hufnagel, L. & Geisel, T. Nature 439,
462–465 (2006).
See Editorial, page 698, and News Feature, page 717.
C. A
. HID
ALG
O &
M. C
. GO
NZ
ALE
Z
Mobile phone records reveal peoples’ movements.
“It’s well known that log–log axes tend to make relationships look straight.” — Andrew Edwards