WITS

The WITSclean command is used to clean and process a CSV file downloaded from WITS into a International Trade Network.

The function requires a number of inputs:
- CSV file name
- Year (as the csv file may contain trade data for multiple years, this therefore allows you to create a ITN for a single year from this CSV file)
- Threshold TRUE/FALSE. TRUE - you want to apply a threshold with a cutoff. FALSE - instead of applying a thrshold you extract the backbone of the network, retaining only statistically signifcant edges.
- Cutoff - for the threshold TRUE this is the cutoff point, for FALSE this is the signficance level for the backbone.

#Example
#Year - 2016
#Threshold - TRUE
#Cutoff - 0.01 (Only ties that are at least 0.01% of world trade are retained)

ITN<-WITSclean("CSV file name.csv",2016,TRUE,0.01)

Reference for extracting the backbone of a network: Serrano, M. Á., Boguñá, M. and Vespignani, A. (2009) Extracting the multiscale backbone of complex weighted networks, Proceedings of the National Academy of Sciences, 106(16), pp. 6483–6488.

UN Comtrade

A similar set of commands are also available for data extracted from UN comtrade. However, this command is apllied to a dataframe, rather than the raw CSV file, so that it can be also used if the user made use of the uncomtradr package to download the data.

library(comtradr)

#Find all the codes associated with "motor vechicle"
auto_codes <- ct_commodity_lookup("motor vehicle",
                                    return_code = FALSE,
                                    return_char = TRUE)

##Download the data for trade in these codes (motor vechicles)
auto_trade<- ct_search(reporters = "All",
                  partners = "All",
                  trade_direction = "imports",
                  start_date = "2016-01-01",
                  end_date = "2016-12-31",
                  commod_codes = auto_codes)

##Alteratively you could just read the CSV file of trade data. Where 
##the CSV file was downloaded directly from UN Comtrade. 
##auto_trade<-read.csv("UNcomtrade file name.csv")

##These results involve multiple codes, so have repeated dyads, for instance, the
##dataframe would contain the trade between US and Canada for code 1, and would
##be present in the data for code 2 etc. 

##The following command aggregates this data, so there would be only one 
##US-Canada dyad for all codes, with the value of the dyad the sum of 
##value for the individual codes
auto_trade_clean<-aggregate(trade_value_usd~reporter_iso+partner_iso, 
                            auto_trade, sum)

##We don't apply this in the WITS command, as the search and download
##functionality already incorpoates this routine. 

library(ITNr)
ITN_auto<-Comtradrclean(auto_trade_clean,2016,
                        TRUE,0.01)

If using the uncomtradr package to download trade data, the concordance package is useful to complement it. The concordance package can be used to translate from different commodity coding systems, which is helpful when using uncomtradr as this only allows you to search using certain commodity coding systems.

library(ITNr)
#Single ITN 2016 
data("ELEnet16")

#List of ITNs 2006 - 2016
data("ELEnetList")

Network Visualisation

The ITNr package contains two functions for plotting the International Trade Networks:
1. Plot set
2. Single Plot

The first produces a panel with four network plots. The purpose of this function for a quick inspection of the network structure, whilst the second plot function provides a better, higher quality plot.

library(ITNr)
data("ELEnet16")

ITNplotset(ELEnet16)

library(ITNr)
data("ELEnet16")

#No labels and node coloured by region
ITN_make_plot(ELEnet16,FALSE,TRUE)

## Loading required package: network

## network: Classes for Relational Data
## Version 1.13.0 created on 2015-08-31.
## copyright (c) 2005, Carter T. Butts, University of California-Irvine
##                     Mark S. Handcock, University of California -- Los Angeles
##                     David R. Hunter, Penn State University
##                     Martina Morris, University of Washington
##                     Skye Bender-deMoll, University of Washington
##  For citation information, type citation("network").
##  Type help("network-package") to get started.

## Loading required package: sna

## Loading required package: statnet.common

## 
## Attaching package: 'statnet.common'

## The following object is masked from 'package:base':
## 
##     order

## sna: Tools for Social Network Analysis
## Version 2.4 created on 2016-07-23.
## copyright (c) 2005, Carter T. Butts, University of California-Irvine
##  For citation information, type citation("sna").
##  Type help(package="sna") to get started.

## Loading required package: scales

Plots

library(ITNr)
data("ELEnet16")

ITNdegdist(ELEnet16)

library(ITNr)
data("ELEnet16")

ITNhistdegdist(ELEnet16)

library(ITNr)
data("ELEnet16")

#Plot is for binary imports and exports
ITNimvex(ELEnet16,FALSE)

library(ITNr)
data("ELEnet16")

ITNcorr(ELEnet16)

library(ITNr)
data("ELEnet16")

region_circle_plot(ELEnet16)

Blockmodel Plot

In addition to calculating the blockmodel membership, we introduce a function to plot the ITN with node coloured by block membership.
You can also set whether you want node labels present (TRUE) or absent (FALSE).

library(ITNr)
data("ELEnet16")

#No labels 
ITNblock_plot(ELEnet16,FALSE)

ERGM & Extensions

Exponential Random Graph models (ERGMs) are generative models that treat network topology (local network substructures) as the response variable. This make them ideal to inform on how and why economic ties occur.

There are a number of variants of the ERGM, and a selection of these can be implemented in R. In this section, we give the very basic exaples of how these can be applied to the ITN, using the network created using the ITNr package. There is a wide range of literature and onlien resoucres, which provide in depth guide on how to implement ERGMs.

When examining the model results, we suggest using the texreg, in particular the screenreg command, as these present the model results in a nice format.

Literature:
Hunter, D. R., Handcock, M. S., Butts, C. T., Goodreau, S. M. and Morris, M. (2008) ergm: A Package to Fit, Simulate and Diagnose Exponential-Family Models for Networks, Journal of statistical software, 24(3), p. nihpa54860.

Lusher, D., Koskinen, J. and Robins, G. (2013) Exponential random graph models for social networks: theory, methods, and applications.

Robins, G., Pattison, P., Kalish, Y. and Lusher, D. (2007) An introduction to exponential random graph (p*) models for social networks, Social Networks, 29(2), pp. 173–191.

Other helpful resources:
-http://michaellevy.name/blog/ERGM-tutorial/
-https://statnet.org/trac/raw-attachment/wiki/Sunbelt2016/ergm_tutorial.html

##Load the data
library(ITNr)
data("ELEnet16")

##Convert igraph object to network object
library(intergraph)
ITN<-asNetwork(ELEnet16)

library(statnet)
##Basic example (not implemented)
ITN.ergm.fit<-ergm(ITN~edges+mutual
                   +gwidegree(0.7,fixed=TRUE)+gwodegree(0.7,fixed=FALSE)
                   +gwesp(0.7,fixed=TRUE)+gwdsp(0.7,fixed=TRUE)
                   +nodematch("region")
                   )

##Examine model results
library(texreg)
screenreg(ITN.ergm.fit)

##Goodness of fit
ITN.ergm.gof<-gof(ITN.ergm.fit)

##Plot goodness of fit
par(mfrow=c(2,2))
plot(ITN.ergm.gof)

##Load data
library(ITNr)
data("ELEnet16")

##Convert ITN to weighted adjacency matrix
library(igraph)
ITNmat<-get.adjacency(ELEnet16,attr="weight") %>% as.matrix()

##Get covariates dataframe
ITN_covariates<-get.data.frame(ELEnet16,what="vertices")

library(GERGM)

##Basic example (not implemented)
formula <- ITNmat ~ edges + mutual(alpha = 0.8)+nodematch("region")+ sender("logGDP") +  receiver("logGDP")  

GERGM.test<- gergm(formula,covariate_data = ITN_covariates,
                   number_of_networks_to_simulate = 40000,
                   thin = 1/100,
                   proposal_variance = 0.05,
                   MCMC_burnin = 10000,
                   seed = 456,
                   convergence_tolerance = 0.5)

Latent Space Models

Below is an example of how to implement a latent space model using latentnet. However, this is a very basic example, see Krivitsky & Handcock (2008) for more detail on implementation of this model.

Reference: Krivitsky, P. N. and Handcock, M. S. (2008) Fitting Latent Cluster Models for Networks with latentnet, Journal of Statistical Software, 24(i05), Available at: http://econpapers.repec.org/article/jssjstsof/24_3ai05.htm.

##Load the data
library(ITNr)
data("ELEnet16")

##Convert igraph object to network object
library(intergraph)
ITN<-asNetwork(ELEnet16)

library(latentnet)
ITN.latentnet<-ergmm(ITN~nodematch("region",diff=TRUE),tofit="mle", verbose=TRUE)

##Summary of the results
library(texreg)
screenreg(ITN.latentnet)

plot(ITN.latentnet,what="mle", pie=TRUE,labels=TRUE)

Preferential Attachment & Fitness Models

The following code provides an example of how the PAFit package can be applied to longitudinal network data - a set of ITNs, one for each year. TO examine whetehr the degree distribution of the networks follwoing a power law or a fitness model.

We use example data from the ITNr package and a command to convert this into a dynamic network object - these commands are described in more detail in the Longitudinal Network Analysis section.

For more details on how to implement the PAFit package and interpret the results see: Pham, T., Sheridan, P. and Shimodaira, H. (2015) PAFit: A Statistical Method for Measuring Preferential Attachment in Temporal Complex Networks, PLOS ONE, 10(9), p. e0137796.

Tutorial:https://cran.r-project.org/web/packages/PAFit/vignettes/Tutorial.pdf

##Load a list of ITNs - this is 11 ITNs, one for each year for 2006-2016
data("ELEnetList")

##Create a dynamic network object for the list of ITNs
ITNdyn<-ITNdynamic(ELEnetList)

##Create a PAFit object from the dynamic network object
ITN_PAFIT<-from_networkDynamic(ITNdyn)

##summary statistics of the temporal network
net_stats <- get_statistics(ITN_PAFIT)
summary(net_stats)

##Estimate the attachment function and node fitness simultaneously.
full_result <- joint_estimate(ITN_PAFIT, net_stats)

##Estimation summary
summary(full_result)

##Plot Attachment
plot(full_result, net_stats, plot = "A")

##Plot Fitness
plot(full_result, net_stats, plot = "f")

Network Autocorrelation Models

Network Autocorrelation models can be implmented using the lnam function in the sna package, or the tnam function in the xergm package. For additional details on how to implement the lnam function function see: https://rdrr.io/cran/sna/man/lnam.html

The tnam is the temporal version of the network autocorrelation model. For a full list of terms available to specify in the tnamsee:https://cran.r-project.org/web/packages/tnam/tnam.pdf and https://www.rdocumentation.org/packages/tnam/versions/1.6.5/topics/tnam

In the context of international trade, these models can be used Smith et al (2016)provide an empirical example, implement the temporal network autocorrelation model to assess how developing efficient trade ties with competitive countries impacts performance in the automotive sector.

Smith, Matthew, Gorgoni, Sara and Cronin, Bruce (2016) The Fragmentation of Production and the Competitiveness of Nations in the Automotive Sector–A Network Approach, Available at: http://economics.uark.edu/conference-files/The-Fragmentation-of-Production-and-the-Competitiveness-of-Nations-in-the-Automotive-Sector.pdf.

Longitudinal Network Analysis

Trade data is availble for decades, therefore longitudnal analysis is also avaiable. Here we present function on how to create a dynamic network object based on a set of (temporal) ITNs (e.g. one ITN for each year).

To create a dynamic network object, we need to specify a list of ITNs. These are usually an ITN for each year, where they are trade in the same product groups across time.

##Load a list of ITNs - this is 11 ITNs, one for each year for 2006-2016
data("ELEnetList")

##Create a dynamic network object for the list of ITNs
ITNdyn<-ITNdynamic(ELEnetList)

## Neither start or onsets specified, assuming start=0
## Onsets and termini not specified, assuming each network in network.list should have a discrete spell of length 1
## Argument base.net not specified, using first element of network.list instead
## Created net.obs.period to describe network
##  Network observation period info:
##   Number of observation spells: 1 
##   Maximal time range observed: 0 until 11 
##   Temporal mode: discrete 
##   Time unit: step 
##   Suggested time increment: 1

##We can also inspect this dynamic network object
head(as.data.frame(ITNdyn))

##   onset terminus tail head onset.censored terminus.censored duration
## 1     0        8   24    2          FALSE             FALSE        8
## 2     9       10   24    2          FALSE             FALSE        1
## 3     0        1   33    2          FALSE             FALSE        1
## 4     2        7   33    2          FALSE             FALSE        5
## 5     8        9   33    2          FALSE             FALSE        1
## 6    10       11   33    2          FALSE             FALSE        1
##   edge.id
## 1       1
## 2       1
## 3       2
## 4       2
## 5       2
## 6       2

Dynamic Network Analysis & Animation

There are many additional packages and functions to analyse this dynamic network object. These are listed at:http://statnet.csde.washington.edu/workshops/SUNBELT/current/ndtv/ndtv_workshop.html#introduction-to-this-workshop

Below is some example code to complement this package and analyse the dynamic network ITN object.

##Create a dynamic network object for the list of ITNs
ITNdyn<-ITNdynamic(ELEnetList)

par(mar=c(1,1,1,1))

filmstrip(ITNdyn,displaylabels=FALSE,vertex.cex=2,
          edge.col="#CCCCCC55",vertex.col='red')

title("filmstrip example")

##Create a dynamic network object for the list of ITNs
ITNdyn<-ITNdynamic(ELEnetList)

render.d3movie(ITNdyn,
               render.par=list(tween.frames=20),
               vertex.cex=0.8,
               vertex.col='red',
               edge.col="#CCCCCC55",
               output.mode = 'htmlWidget')

##Commands to save animation as GIF (to insert in presentations)
saveGIF(render.animation(ITNdyn,
                       render.par=list(tween.frames=20),
                       vertex.cex=0.8,vertex.col='red',
                       edge.col="#CCCCCC55"),movie.name = "Example.gif")

##Create a dynamic network object for the list of ITNs
ITNdyn<-ITNdynamic(ELEnetList)

##Timeline Plots
timeline(ITNdyn)
timeline(ITNdyn,plot.edge.spells=FALSE)

#tsna - Temporal Social Network Analysis
library(tsna)
tEdgeFormation(ITNdyn)
plot(tEdgeFormation(ITNdyn))

tSnaStats(ITNdyn,'gtrans')
plot(tSnaStats(ITNdyn,'gtrans'))