Germs, Social networks and Growth Introduction Related Work Fixed Network Model Evolutionary Network Model Calibration Estimation Conclusion Germs, Social Networks and Growth Alessandra Fogli 1 Laura Veldkamp 2 1 University of Minnesota 2 NYU Stern Summer 2012
47
Embed
Germs, Social Networks and Growth...Germs, Social networks and Growth Introduction Related Work Fixed Network Model Evolutionary Network Model Calibration Estimation Conclusion Social
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
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Germs, Social Networks and Growth
Alessandra Fogli1 Laura Veldkamp2
1University of Minnesota
2NYU Stern
Summer 2012
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Motivation
Large literature explaining differences in output acrosscountries:
Political, legal, financial institutionsGeography, climate and factor endowmentsTechnological progress and technology diffusion
All abstract from the potential role of “social structure”
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Social Structure
Definition: The pattern of social ties between people in aneconomy; a social network.
Our main questions:Do differences across countries in social structurematter for macroeconomic outcomes?How might they matter?Where do they come from?
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Social Networks
A social network: the set of all individuals in theeconomy (“nodes”) and their relationships (“ties” or“edges”).How individuals make friends determines the socialnetwork. Aggregate features of this network constitutesocial structure.The structure of a network affects the speed oftechnology diffusion: differences in social structuretranslate into differences in the speed of diffusion ofideas in the economy.
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Key feature: Collectivism in networks
Dimension of variation: tendency to create tightly knitgroups (collectives).Collective: a set of 3 mutually connected nodes.Example: How collectives in networks can affectdiffusion.
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
1
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
2
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
3
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
4
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
5
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
6
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
7
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
8
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
9
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
10 - DONE
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Key Ideas
If some social structures slow down the diffusion ofideas,
Why do they emerge in the first place?
Two factors that affect development and spread throughhuman contact:
Germs and Ideas
Social networks that inhibit the diffusion of ideas alsoprotect people from diseases.
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Outline
Related workCompare diffusion in 2 exogenous networks: 1individualist and 1 collectivist (isolate this dimension)Endogenize the network: Evolutionary model with highand low disease explains why networks form.Data: Collectivism, pathogens, tech diffusionIV estimation: Use the difference between sociallytransmittable and zoonotic disease to instrument forsocial structure. Estimate effect on technology.
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Related work
Technology diffusion: Lucas and Moll (2011), Perla andTonetti (2011) use random matching.Technology spillovers: Eeckhout and Jovanovic (2002).Our network puts more structure on matches orspillovers.Political structure and growth: Acemoglu, Johnson andRobinson (2002). Related empirical strategy.Collectivism, culture and norms: Gorodnichenko andRoland (2011), Bisin and Verdier (2000,01), Fernandezand Fogli (2005), Greif (1994).
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
While mass media play a major role in alertingindividuals to the possibility of an innovation, itseems to be personal contact that is most relevantin leading to its adoption. Thus, the diffusion of aninnovation becomes a process formally akin to thespread of an infectious disease.
(Kenneth Arrow, 1969 AEA presidential address)
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Fixed Network Model
Time discrete and infinite.N agents, located on a circle, are indexed by theirlocation i. 2 states:
health status: healthy (ψi (t) = 0) or sick (ψi (t) = 1)technology level: Ai (t)
Network: A matrix N of 1’s and 0’s. nij = 1 means that i ,j connected.A collective: When friends i and j have a mutual friendk . nij = nik = njk = 1
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Fixed Network Model: Disease
Healthy agents produce Ait . They survive to the nextperiod with probability ξ.Sick agents have zero productivity (Ai(t) = 0) and dieat end of period.Initial fraction S of sick people.Each sick person transmits the disease to each friendwith probability π.An agent who dies is replaced. The new agent i , bornat date t , has same network and Ai(t−1) = 0.
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Fixed Network Model: Technology
Technology advances in 2 ways:Poisson arrival of new ideas. At the start of eachperiod, with probability λ agent j advances histechnology by one step:
ln(Aj(t + 1)) = ln(Aj(t)) + δ
Agents can also learn from others in their networks. Ifperson j is connected to person k and Aj(t) > Ak (t)then with probability φ, k will learn what j knows:
Ak (t + 1)) = Aj(t)
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Fixed Network Model:
We compare diffusion in 2 networks.Collectivist network (N1): Hold the number ofconnections fixed (4). Everyone connected. Maximizecollectives.Individualist network (N2): The minimal deviationfrom collectivist network that achieves zero collectives.
Key result: Collectives slow diffusionCollectives increase network diameter.Agents are more likely to infect friends that havealready been infected by other friends.
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Fixed Network Model: Diffusion Results
Let Ψj(t) = min{s ≥ t : ψj(s) = 1} andαj(t) = min{s ≥ t : Aj(s) > A(t)}.
Result
If π = 1 and∑
j ψj(0) = 1, then the average lifetimeEj [Ψj(0)] is longer in the collectivist network (1) than in theindividualist network (2).
If φ = 1, then the average discovery time Ej [αj(0)] is slowerin the collectivist network (1) than in the individualistnetwork (2).
Collectivist networks prolong lifetimes, but slow technologydiffusion.
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Fixed Network Model: Technology FrontierResults
Result
Suppose that at t, a collectivist network (N1) and anindividualist network (N2) have the same Aj(t) ∀j . Then theprobability that the next new idea arrival will increase thetechnological frontier is larger in N2 than N1.
This is why an individualist network can achieve a higherrate of growth and a higher income in the long-run.
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
An evolutionary model of network formation
Same structure as before, exceptIndividuals can be of two types: collectivist τj(t) = 0and individualist τj(t) = 1.
Collectivist in location j is linked to j − 1, j + 1 and j + 2Individualist in location j is linked to j − 1, j + 1 and j + 4They can both be linked to j − 2 and/or j − 4 dependingon types of agents in those locations.
When agents die (from disease or old age), new agentsinherit the type and technology of the friend in theirparent’s network with highest A. ⇒ Successful typesare passed on.
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Long run results
ResultWith probability 1, the network becomes homogeneous: ∃Ts.t. τj(t) = τk (t) ∀k and ∀t > T .
ResultWith probability 1, the disease dies out: ∃T s.t. ψj(t) = 0 ∀jand ∀t > T .
If individualist= 0⇒ Stay collectivist foreverIf individualist> 0⇒ Likely converge to individualist
(learns faster and pass on type more often but not certainb/c random death)
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Calibrate the model
Does higher disease⇒ collectivism?How much can networks affect growth?
Parameter Value TargetInitial disease E [ψj (0)] 0.5% TB death rateprevalence in ChinaDisease transmission π 32% Disease disappears inprobability 150 years (indiv avg)Innovation δ 30% 2.6% growth rate inproductivity increase individualist countryTechnology transfer φ 50% Half-diffusion inprobability 20 years (Comin et. al. ’06)
Figure: How disease and technology spread through networks.
The darkest boxes indicate individuals who acquired the disease inperiod t and therefore have zero time-t productivity. Warmer colors
indicate higher levels of technology. Note: these are not calibrated levelsof disease or tech arrival.
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Calibration results: Comparing fixed networks
0 50 100 150 200 2500
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
Period
Collectivist Network
Average Technology*10−4
Disease Rate
0 50 100 150 200 2500
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
Period
Individualist Network
Average Technology*10−4
Disease Rate
Figure: Average disease prevalence and productivity
Avg growth rate in individualist network: 2.6%Avg growth rate in collectivist network: 2.0%
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Numerical example: Network Evolution
Long run network structure depends on whether disease orindividualist trait dies out first.Example with 1 run:
0 50 100 150 200 250 300 350 4000
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16High Disease
% Individualist% Infected
0 50 100 150 200 250 300 350 4000
0.02
0.04
0.06
0.08
0.1
0.12Low Disease
% Individualist% Infected
Figure: Long run network structure
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Calibration results: Disease and networkevolution
Evolutionary model: Higher disease prevalence makescollectivist network more likely.
0 50 100 150 200 2500
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8Low Initial Disease
% Disease Extinct% Indivs Extinct
0 50 100 150 200 2500
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8High Initial Disease
% Disease Extinct% Indivs Extinct
The probability that the economy converges to a zero-disease or purelycollectivist steady-state at each date.
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Theory summary
Initial differences in disease govern social structure,which persists even after diseases disappear.The collectivist economy grows .6% less per year. Alarge quantitative effect over the long-run.Next step: Can we find causal evidence of an effect ofsocial structure on income? How big?
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Empirical strategy
Instrument suggested by theory: historical prevalenceof pathogens. But pathogens depend on technologyand disease directly affects productivity.
A = α1 + α2S + ε
S = γ1 + γ2A + γ3GH + γ4GZ + η
Gi = δA + εi iε{H,Z}
Problem: Gi ’s show up in residual ε.Identifying assumption: E [(GH −GZ )ε] = 0Our instrument: human−zoonotic pathogens.Idea: Both have the same relationship with A, but onlyhuman-to-human pathogens affect S.
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Empirical measures
Social structureindividualism index: constructed by sociologists tomeasure dependency of individuals from the group(varies between 0 and 100).
Innovationspeed of technology diffusion from Comin and Mestieri(2012).
Germsprevalence of 9 of the deadliest communicablediseases: leishmanias, leprosy, trypanosomes, malaria,schistosomes, filariae, dengue, typhus and tuberculosis,from 1930 atlases of infectious diseases (on a 4 pointsscale).
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Measuring Collectivism
Two observable characteristics of collectivism1 An aversion to severing social ties
Difficult to sustain collectives with changing ties.2 Cooperation and social influence
Common friends enforce cooperation and social norms.
Key factors that determine Hofstede’s individualismindex
1 Group cooperation (C)2 Importance of freedom (I)3 Importance of job satisfaction and location(I)
Collectivists don’t change jobs or location much.
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Constructing Disease Instruments
Instrument 1: Difference in reservoir (diff_res)1 Reservoir is the long-term host of a pathogen.2 Human: leprosy, filariae3 Zoonotic: schistosomes, typhus4 Multi-host: leishmanias, trypanosomes, malaria,
All variables are included in the second stage and first stage regressions.The other instruments are diff_res_std, pronoun, the fraction of
English-speakers and a constant.
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Estimating Effects on Output
Dependent var: Solow Residual Output per capitaInstruments: diff_res_std diff_res_std
pronoun, eng pronoun, engIndividualism 0.99 2.10
(0.40) (0.45)Over-ID p-val 0.78 0.87
Accept AcceptR2 0.20 0.42N 58 59.
1 std dev increase in individualism (23.0) results in 23 higherSolow residual (23% of US level) and 48 higher output per capita(48% of US level). Solow residual and output per capita come from the
Penn World Tables mark 5.6. All estimates are significant at 5% level.
Germs, Socialnetworks and
Growth
Introduction
Related Work
Fixed NetworkModel
EvolutionaryNetworkModel
Calibration
Estimation
Conclusion
Conclusions
If ideas and germs spread in similar ways, diseaseprevalence can rationalize social networks that inhibittechnology diffusion and growth.Differences in social networks persist, even afterdisease disappears. Large income effects over time.IV analysis finds evidence of the effect of socialnetworks on technology.More generally we offer a theory of endogenous socialinstitutions and show how to measure and test for theireffects.