remx

A package with meta-analytic approximations for relational event models.

This package supports mixed-effects models, data streams, and large networks.

Installation

install.packages("devtools")
devtools::install_github("TilburgNetworkGroup/remify") 
devtools::install_github("TilburgNetworkGroup/remstimate")
devtools::install_github("TilburgNetworkGroup/remstats")
devtools::install_github("TilburgNetworkGroup/remx")

Mixed-Effect Models

In this case, there are multiple independent networks that need to be analized using a multilevel model.

The remx function is the main function of the package. It fits all multilevel models available.

Tie-oriented Model

#----------------------------#
#     Tie-Oriented model     #
#----------------------------#

#Loading libraries
library(remx)
library(remstats)

#Loading the data
edgelist <- networks

#Computing statistics
effects <- ~ remstats::inertia(scaling = "std") + remstats::reciprocity(scaling = "std")
rehObj <- lapply(1:length(edgelist), function(x) remify::remify(edgelist[[x]], model = "tie"))
stats <- lapply(1:length(edgelist), function(x) remstats::tomstats(effects, rehObj[[x]]))

#Running the model

fit <- remx(reh = rehObj,
            statistics = stats,
            random = c("baseline", "inertia", "reciprocity"))

print(fit)

Actor-oriented Model

#----------------------------#
#    Actor-Oriented model    #
#----------------------------#

#Computing statistics
rehObj <- lapply(1:length(edgelist), function(x) remify::remify(edgelist[[x]], model = "actor"))
sender_effects <- ~ indegreeSender(scaling = "std")
receiver_effects <- ~ indegreeReceiver(scaling = "std") + outdegreeReceiver(scaling = "std")
statistics <- lapply(1:length(edgelist), function(x) remstats::aomstats(rehObj[[x]], sender_effects, receiver_effects))

fit <- remx(reh = rehObj,
            statistics = statistics,
            random_sender = c("baseline", "indegreeSender"),
            random_receiver = c("indegreeReceiver", "outdegreeReceiver"))

print(fit)

Extracting model parameters

These functions are inspired by the functions contained in the lme4 package. The can be used to extract the parameters for models fitted using the remx function.

Random effects

ranef_metarem(fit)

Fixed effects

fixef_metarem(fit)

Random-effect covariance matrix

VarCov_metarem(fit)

Relational-event Network Data Streams

In this case, we have a relational-event network that is augmented with additional events over time. Every new batch of events will be used to update the model estimates.

Computing network statistics

We need to use the arguments start and stop from remstats in order to compute the statistics for a specific portion of the event sequence. For instance, if we wish to fit the model in batches of 500, we need to declare start = 1 and stop = 500, then start = 501 and stop = 1000, and so on.

#Now let's simulate a network
edgelist <- stream$edgelist

library(remstats)

#We will devide the network in 10 batches of 500
events <- seq(1, 5001, by = 500)

#Declaring which effects we want remstats to compute
effects <- ~ remstats::inertia(scaling = "std") + 
  remstats::reciprocity(scaling = "std") + 
  remstats::indegreeSender(scaling = "std") +
  remstats::outdegreeSender(scaling = "std")

#Getting the remify object
rehObj <- remify::remify(edgelist, model = "tie", actors = stream$actors)

data <- vector("list")

for(i in 2:length(events)){
  #Computing statistics
  stats <- remstats::tomstats(effects, rehObj, start = events[i-1], stop = events[i]-1)
  edl <- edgelist[events[i-1]:(events[i]-1),]
  #Every piece needs to be stored a in a list with edgelist and statistics
  data[[i-1]] <- list(edgelist = edl,
                      reh = remify::remify(edl, model = "tie", actors = stream$actors),
                      statistics = stats)
}

Fitting the model

#Let's compute the effects for the first 7 batches of the networks
fit <- strem(data[1:7])

#printing the parameters
print(fit)

Plotting trends of the estimates across batches

The package also contains a generic plot function, that can be used to plot the trends, along with 95% intervals, of the estimated effects across each batch. This function only works for models fitted with the function strem.

#Getting a plot of the estimates with confidence intervals
plot(fit)

Updating the model with new batches of events

We can update our estimates with new batches by simply passing a model previously fitted with the strem function to the argument model and use the argument update = TRUE.

#Now we can update the model with the remaining 3 batches
fit <- strem(data[8:10], update = T, model = fit)

#printing the parameters
print(fit)

Citation

Vieira, F., Leenders, R. & Mulder, J. Fast meta-analytic approximations for relational event models: applications to data streams and multilevel data. J Comput Soc Sc 7, 1823–1859 (2024). https://doi.org/10.1007/s42001-024-00290-7

@article{vieira2024fast,
  title={Fast meta-analytic approximations for relational event models: applications to data streams and multilevel data},
  author={Vieira, Fabio and Leenders, Roger and Mulder, Joris},
  journal={Journal of Computational Social Science},
  pages={1--37},
  year={2024},
  publisher={Springer}
}

Acknowledgement

The development of this package was supported by a Vidi Grant (452-17-006) awarded by the Netherlands Organization for Scientific Research (NWO) Grant and an ERC Starting Grant (758791).