networkDynamicData (version 0.1.0)

concurrencyComparisonNets: A synthetic dataset of three simulated dynamic networks with epidemic spread.

Description

Three single-mode undirected dynamic networks with an infection started from a single seed, The networks were simulated using the tergm and EpiModel packages. All three networks have the same size, relationship duration distribution and cross-sectional mean degree, but different cross-sectional degree distributions. They are intended as examples for illustrating and comparing the effects of concurrent overlapping partnerships on the connectivity and dynamic transmission potential of networks.

Usage

data(concurrencyComparisonNets)

source

http://statnet.org

Details

Each network has the following shared characteristics: 1000 nodes, 100 timesteps, a cross-sectional mean degree that varies stochastically around 0.75, and an exponential relationship duration distribution with a mean of 25 timesteps (due to censoring effects, the naive mean duration calculation using all observed partnerships will be around 20). The only difference in the three networks is the cross-sectional degree distribution, varying from Bernoulli (monog) to Poisson (base), which represent a range from strict serial monogamy in partnerships, to the levels of concurrency that would be present if partnerships are formed independently, without regard for any existing partnerships (an Erdos-Renyi graph). This is accomplished by modifying the the formation model of the STERGM used to simulate edge dynamics (see accompanying code for details).

After simulating the dynamic network, a trivial disease simulation is implemented from a single seed in each network, with transmission probability set to 1.0. For each discordant partnership formed, transmission is therefore guaranteed in one timestep, and the infections trace out the size of a forward-reachable component in each network. Note that because the dynamic network is simulated in its entirety first, this implies the partnership formation/dissolution process is independent of the disease state of the node and the network.

Each network has a dynamic 'status' nodal attribute indicating the infection status of each node at each timestep in each network. Comparison of the prevalence and trajectories of the status variable provide insight into the impact of concurrent partnerships on network connectivity and transmission potential.

The networks were simulated using the EpiModel package and the code below.

Terms and Conditions{ The concurrencyComparisonNets data are provided under the tergms of the Creative Commons Attribution 3.0 License: http://creativecommons.org/licenses/by/3.0/us/

Please cite the dataset authors and the networkDynamicData package (citation(package='networkDynamicData')) with any redistribution or published use of this data.

}

References

Morris M., Kurth A., Hamilton D.T., Moody J., and Wakefield S., for The Network Modeling Group (2009) "Concurrent Partnerships and HIV Prevalence Disparities by Race: Linking Science and Public Health Practice" American Journal of Public Health 1023-1031, Vol 99, No. 6

Jenness S, Goodreau S, Wang L and Morris M (2014). EpiModel: Mathematical Modeling of Infectious Disease. The Statnet Project (http://www.statnet.org). R package version 0.95, CRAN.R-project.org/package=EpiModel.

Examples

data(concurrencyComparisonNets)

# compare plots of each network at time 50
plot(network.extract(base,at=50),vertex.cex=0.5,edge.lwd=2)
plot(network.extract(monog,at=50),vertex.cex=0.5,edge.lwd=2)
plot(network.extract(middle,at=50),vertex.cex=0.5,edge.lwd=2)

# compare mean degree. These are at ~20, due to censoring
mean(as.data.frame(base)$duration)
mean(as.data.frame(middle)$duration)
mean(as.data.frame(monog)$duration)

# compare infection time series
plot(sapply(1:100,function(t){
    sum(get.vertex.attribute.active(base,'status',at=t))
  }),col='black',xlab='time step', ylab='# infected'
)
points(sapply(1:100,function(t){
     sum(get.vertex.attribute.active(monog,'status',at=t))
   }),col='blue')
points(sapply(1:100,function(t){
     sum(get.vertex.attribute.active(conc,'status',at=t))
   }),col='red')   

## The code below was used to generate the three example networks using EpiModel
## note that 'conc' is renamed 'middle' and the networks have 
## some attached simulation control variables deleted before 
## being saved as the datasets 

library(EpiModel)

# === example network parameters setup ===

params.base = list(
  sim.length = 100,
  size = 1000,
  mean.deg = .75,
  mean.rel.dur = 25,
  
  net = network.initialize(1000, directed = F),
  formation = ~edges,
  dissolution = ~offset(edges)
)

params.conc = list(
  sim.length = 100,
  size = 1000,
  mean.deg = .75,
  mean.rel.dur = 25,
  
  net = network.initialize(1000, directed = F),
  formation = ~edges+concurrent,
  dissolution = ~offset(edges),
  targets = 90  # concurrent = 90
)

params.monog = list(
  sim.length = 100,
  size = 1000,
  mean.deg = .75,
  mean.rel.dur = 25,
  
  net = network.initialize(1000, directed = F),
  formation = ~edges+concurrent,
  dissolution = ~offset(edges),
  targets = 0  # concurrent = 0
)


# === function for estimating stergm, simulating network, and simulating epidemic ===

net.init = function(params, nsims, adjust=F) {
  for (name in names(params)) assign(name, params[[name]])
  
  message('network init')
  
  # generate initial network (defaults if not specified in params)
  if (!exists('net', inherits=F)) {
    net <- network.initialize(size,directed=F)
    net %v% 'race' <- rep(0:1, each=250)
  }
  if (!exists('formation', inherits=F)) {
    formation = ~edges
  }
  if (!exists('dissolution', inherits=F)) {
    dissolution = ~offset(edges)
  }
  
  if (!is.null(mean.deg)) {
    target.edges <- size/2 * mean.deg
    density = target.edges / choose(size,2)
  } else {
    target.edges <- round(density*choose(size, 2))
  }
  print(target.edges)
  
  # cludge to fix the monogamy bias in simulate
  if (adjust) target.edges = target.edges*adjust
  
  # target stats that does not include edges
  if (exists('targets', inherits=F)) {
    target.stats = c(target.edges, targets)
  } else {
    target.stats = target.edges
  }
  
  coef.diss <- dissolution.coefs(dissolution, mean.rel.dur)
  
  # estimate the stergm
  
  net.est = epiNet.est(net, formation, dissolution, target.stats, coef.diss
                       ,control=control.ergm(MCMLE.maxit=200))
  
  stats.form = update(formation, ~.+concurrent)
  
  # simulate the dynamic network
  
  net.sim = epiNet.simNet(net.est, nsteps = sim.length, nsims=nsims, stats.form=stats.form,
                          control = control.simulate.network(MCMC.burnin.add=10))
  
  # simulate the epidemic
  
  trans.sim = epiNet.simTrans(net.sim, "SI", vital=FALSE, i.num=1, trans.rate=1, tea=TRUE)
  
  print(summary(net.sim$stats[[1]]))
  plot(net.sim$stats[[1]][,'edges'], ylab='edges', xlab='time')
  
  return(trans.sim)
}


# === simulate example networks ===

base.sims = net.init(params.base, 1)
base = base.sims$network[[1]]

conc.sims = net.init(params.conc, 1)
conc = conc.sims$network[[1]]

monog.sims = net.init(params.monog, 1)
monog = monog.sims$network[[1]]