desktop 🠖 tutorials
Working with Network Canvas data in R
Summary:
This tutorial will show you how to work with Network Canvas data in the R statistics environment.
Prerequisites:
In order to follow along with this tutorial you should: (1) Have an understanding of the R environment. (2) Have a working installation of R studio, or similar, in order to enter commands.
Duration:
30 minutes
This tutorial provides basic examples of import and analysis in R using data from the Network Canvas software. The example will use simulated data from the following network canvas protocol file which can be opened in the Network Canvas Interviewer or Architect apps. The data used in this tutorial can also be found here.
Find the github repository for this tutorial here: https://github.com/complexdatacollective/working-with-data
Export data from Interviewer or Fresco
Interviewer
The first step in exporting data from Interviewer is to navigate to the Manage and Export Session data in the Interviewer app.
Next, you’ll indicate which sessions you want to export by selecting the appropriate cards and clicking “Export Selected To File.”
Next, we’ll confirm export options. We will focus on the CSV export in this tutorial. I also leave the “Merge Sessions” option unselected because I prefer to aggregate session data manually in R, as I will demonstrate below.
Finally, after a short wait, we simply select where we would like the zip file containing all our data to be saved on our computer. We will need to unzip the file before starting our data analysis described below
Fresco
For information on how to export data in Fresco, see using Fresco.
Import and clean the data
Next we want to load the data into R. We use the egor to do this, using examples derived from an egor vignette.
# Load egor
library(egor)
library(sna)
library(ggplot2)
# This assumes the the data was unzipped to the "networkCanvasExport" folder that is located in the same folder as this script
# If the you have data somewhere else, set that folder path here:
folderPath <- paste0(getwd(),'/networkCanvasExport/')
# Read each type of file into list, combine into single data frame
alterData <- folderPath %>%
list.files(full.names=TRUE,pattern="attributeList_Person.csv") %>%
lapply(read.csv) %>%
bind_rows()
edgelistData_Close <- folderPath %>%
list.files(full.names=TRUE,pattern="edgeList_Close.csv") %>%
lapply(read.csv) %>%
bind_rows()
edgelistData_SexTie <- folderPath %>%
list.files(full.names=TRUE,pattern="edgeList_SexTie.csv") %>%
lapply(read.csv) %>%
bind_rows()
edgelistData_DrugTie <- folderPath %>%
list.files(full.names=TRUE,pattern="edgeList_DrugTie.csv") %>%
lapply(read.csv) %>%
bind_rows()
egoData <- folderPath %>%
list.files(full.names=TRUE,pattern="ego.csv") %>%
lapply(read.csv) %>%
bind_rows()
Now we can examine if the data looks like it should from the alter file. As you’ll see, the alter data file contains a unique alter identifier (i.e., “networkCanvasUUID”) as well as a unique ego identifier (i.e., “networkCanvasEgoUUID”) which will both be used to define the egor object.
head(alterData[,c("nodeID","networkCanvasEgoUUID","networkCanvasUUID","Close","Drugs","Sex","Age")])
## nodeID networkCanvasEgoUUID
## 1 1 1b915958-a970-4742-9f98-154a5b97c3e9
## 2 2 1b915958-a970-4742-9f98-154a5b97c3e9
## 3 3 1b915958-a970-4742-9f98-154a5b97c3e9
## 4 4 1b915958-a970-4742-9f98-154a5b97c3e9
## 5 5 1b915958-a970-4742-9f98-154a5b97c3e9
## 6 6 1b915958-a970-4742-9f98-154a5b97c3e9
## networkCanvasUUID Close Drugs Sex Age
## 1 38eea03a-de1e-4a0b-bd22-9424cd1d9011 true false false 28
## 2 21fc603a-34af-42eb-a317-14b9f04d811f true false true 91
## 3 b22bcff1-e4b9-4aa8-b4cd-0667a1f19ea9 false false true 90
## 4 df31eaa1-bc6d-4213-a25e-61536391a23f false false false 51
## 5 c9a1103c-2246-427c-bd7b-f11ea1bf242f true true true 82
## 6 c202aa1a-9d24-48b1-82ea-6dbc5e9580f7 true true false 56
Recoding categorical variables
Categorical variables from Network Canvas export as one column per category with a Boolean value for each possible category. However, researchers frequently want to convert these variables to a single factor column in R. This helper function can help you achieve that.
catToFactor <- function(dataframe,variableName) {
fullVariableName <- paste0(variableName,"_")
catVariables <- grep(fullVariableName, names(dataframe), value=TRUE)
# Check if variable exists
if (identical(catVariables, character(0))){
stop(paste0("Cannot find variable named -",variableName,"- in the data"))
# Check if "true" in multiple columns of a single row
} else if (sum(apply(dataframe[,catVariables], 1, function(x) sum(x %in% "true")>1))>0) {
stop(paste0("Your variable -",variableName," - appears to take multiple values.")) }
catValues <- sub(paste0('.*',fullVariableName), '', catVariables)
factorVariable <- c()
for(i in 1:length(catVariables)){
factorVariable[dataframe[catVariables[i]]=="TRUE"] <- catValues[i]
}
return(factor(factorVariable,levels=catValues))
}
Using this function, we can recode three variables from the current protocol into factors
# List of categorical variables in our protocol to convert into factors
categoricalVariablesList <- list('Gender','Race','SexOrient')
# Iterate the list and call our catToFactor function, assigning the result to a new column in our dataframe
for (variable in categoricalVariablesList) {
alterData[variable] <- catToFactor(alterData, variable)
}
Miscellaneous data cleaning
We’ll want to do a little data cleaning by contact frequency and counting the number of substances participants reported using, before loading it into an egor object.
alterData$ContactFreq[alterData$ContactFreq=="Less_than_weekly"] <- "Less than \n weekly"
countTrues <- function(vector) {
length(which(vector=="true"))
}
drugNames <-c("MarijuanaUsed","CocaineUsed","HeroinUsed","PainkillersUsed","PoppersUsed")
egoData$numberDrugs <- apply(egoData[,c(drugNames)],FUN=countTrues,MARGIN=1)
Create an egor object
After loading and cleaning the data, we still need to make it into an egor object. Although we have 3 data files with edgelist data, we’ll be focusing on “Close” social ties for the remainder of this tutorial
# Load the file into R
egorNetworkCanvas <- egor(alters = alterData,
egos = egoData,
aaties = edgelistData_Close,
ID.vars = list(
ego = "networkCanvasEgoUUID",
alter = "networkCanvasUUID",
source = "networkCanvasSourceUUID",
target = "networkCanvasTargetUUID"))
Data visualization
Lets start with a simple visualization of one ego network. To do this we’ll first convert it to a ‘network’ object and use the gplot function from the sna package. This visualization shows the new node labels and colors each node by the contact frequency with the participant. We also layout the figure with the final coordinates from the sociogram stage.
oneEgoNet <- as_network(egorNetworkCanvas)[[9]]
oneEgoNet%v%"vertex.names" <- oneEgoNet%v%"name"
colorScheme <- c( "#CC6677", "#117733", "#AA4499",
"#6699CC")
# A little recoding to get a color for each frequency
nodeColors <- ifelse(oneEgoNet%v%"ContactFreq"=="Daily",colorScheme[1],
ifelse(oneEgoNet%v%"ContactFreq"=="Weekly",colorScheme[2],
ifelse(oneEgoNet%v%"ContactFreq"=="Less than \n weekly",colorScheme[3],
colorScheme[4])))
gplot(oneEgoNet,
usearrows = FALSE,
label = oneEgoNet%v%"name",
displaylabels = TRUE,
vertex.col=nodeColors,
edge.col="gray",
coord = matrix(c(as.numeric(oneEgoNet%v%"Cords_x"),
-as.numeric(oneEgoNet%v%"Cords_y")),
nrow=length(unique(oneEgoNet%v%"name")),
ncol=2))
As you can see, this only shows a single egocentric network. However, the egor package has several functions that facilitate comparison of networks across ego nets. For example, here is a visualization showing each ego net with nodes location being dependent on their frequency of contact with the participant and their status as a drug partner (i.e., true/false).
# Make a visualization displaying both frequency of communication and family member status
plot(egorNetworkCanvas, venn_var = "Drugs",
pie_var = "ContactFreq",vertex_label_var="nodeID",
type = "egogram")
Data analysis
The egor package has numerous functions that help with basic data analysis of ego networks. For example, the summary function provides an overview of all ego networks in the egor while the ego_density functions provides the density for each participant’s network.
summary(egorNetworkCanvas)
## 9 Egos/ Ego Networks
## 91 Alters
## Min. Netsize 3
## Average Netsize 10.1111111111111
## Max. Netsize 20
## Average Density 0.385856531573766
## Alter survey design:
## Maximum nominations: Inf
ego_density(egorNetworkCanvas)
## # A tibble: 9 × 2
## .egoID density
## <chr> <dbl>
## 1 1b915958-a970-4742-9f98-154a5b97c3e9 0.619
## 2 102783ac-c48d-4953-97e6-3d84f39fd5d5 0.214
## 3 51ce5e11-9aed-4b07-8ccf-1e96f7194ba6 0.0256
## 4 ef3aeefe-1af9-4fd4-9e75-318eb329778f 0.105
## 5 aae09896-5fc5-4295-8349-58045da6e86a 1
## 6 2944b829-ef9d-446e-b592-6fbf57657743 0.00526
## 7 79cee692-a513-408d-9701-0efe078b0240 0.218
## 8 1c6dd8fd-2b2f-4139-80bc-63c7f34a5ec1 1
## 9 0ce518be-d845-4439-8e3a-bd6b86e7cf40 0.286
We can also use a traditional package, such as sna, to look at these networks by applying functions (i.e., lapply) to each of these networks and aggregating the results. For example, here we first make a simple histogram of alter degrees across all ego networks.
networkNetworkCanvas <- as_network(egorNetworkCanvas)
histData <- networkNetworkCanvas %>%
lapply(degree,cmode="indegree") %>%
unlist(recursive = FALSE) %>%
as.data.frame()
histData$degree <- as.numeric(histData$".")
ggplot(histData, aes(x=degree)) +
geom_histogram(color="black", fill="white",bins=5) +
theme_classic()
Finally, we often want to examine how an ego attribute may be associated with ego network characteristics. In this example we look at the association between the number of substances a participant reported using and the density of their ego network.
ego_density(egorNetworkCanvas) %>%
full_join(egorNetworkCanvas$ego,by=".egoID") %>%
ggplot(aes(x = numberDrugs, y = density)) +
geom_point(size=5) +
geom_text(label=egorNetworkCanvas$ego$networkCanvasCaseID, aes(vjust=c(-1.5))) +
ylim(0,1.05) + theme_classic()
