Outdegree producing NA about tnam HOT 3 CLOSED

Can you post a minimal reproducible example? Thanks.

from tnam.

#I'm not sure about minimal but here goes...

graphoutput.txt

library(igraph)
library(tnam)

#Edit the file to be .csv and then edit this line to read it in
dat <- read.csv("PATH/graphoutput.csv", header=FALSE, sep=",")

dat <- dat[,1:110]
g <- graph_from_adjacency_matrix(data.matrix(dat), mode=c("directed"), weighted=TRUE)

g <- delete_vertex_attr(g, "name")

x <- length(degree(g))

#store one copy the network variable here: 'net1 <- as.integer...' in lab notes
net1 <- array(as.integer(as_adjacency_matrix(g, attr="weight", sparse=FALSE)), dim=c(x,x)) #Just the factories

mode(net1) <- "integer"

V(g)$diffusion=0
good <- 5 #Number of good buyers
s <- c(1:5)#list of the good buyers
V(g)$diffusion[s] <- 1

#store one copy of each covariate 'wt1 <- c(...)' in lab notes
wt <- c() #Total weight
bc <- c() #Buyer connections
sn <- c() #Connected to a seed node
sn2 <- c() #Connected to a node connected to a seed node
for (i in 1:x){
wt <- c(wt, sum(E(g)[from(i)]$weight)) #total weight out as a measure of factory size; doesn't change over time
bc <- c(bc, sum(neighbors(g,i,mode=c("out")) %in% s)) #Number of buyers a factory is connected to
if (sum(neighbors(g,i,mode=c("out")) %in% which(V(g)$diffusion == 1)) > 0){ #Is the node connected to a seed?
sn <- c(sn, 1)
}else{
sn <- c(sn, 0)
}
}

for (i in 1:x){
if (sum(neighbors(g,i,mode=c("out")) %in% which(sn==1)) > 0){
sn2 <- c(sn2, 1)
}else{
sn2 <- c(sn2, 0)
}
}

num_edge <- length(E(g))

goes <- 8
diffs <- matrix(nrow=x, ncol=goes)

for (i in 1:goes){
for (edg in sample(1:num_edge, num_edge, replace=FALSE)){
buyer <- ends(g, E(g)[edg])[,2]
seller <- ends(g, E(g)[edg])[,1]

if (V(g)$diffusion[buyer] == 1){
#in to buyer
into <- sum(E(g)[to(buyer)]$weight)
#out from neighbour =
outof <- sum(E(g)[from(seller)]$weight)

topline <- E(g)[seller %--% buyer]$weight

if (topline/outof > topline/into){V(g)$diffusion[seller] = 1}
}
}
#store one column of the diffusion dependent variable here
diffs[,i] <- V(g)$diffusion[1:x]
}

colnames(diffs) <- c("t1", "t2", "t3", "t4", "t5", "t6", "t7", "t8")
rownames(diffs) <- c(1:x)
mode(diffs) <- "integer"
diffs <- as.data.frame(diffs)

tnet <- list(net1, net1, net1, net1, net1, net1, net1, net1)
names(tnet) <- c("t1", "t2", "t3", "t4", "t5", "t6", "t7", "t8")
#Do the names of all variables:
for (i in 1:length(tnet)) {
rownames(tnet[[i]]) <- c(1:x)
}
names(wt) <- c(1:x)
names(bc) <- c(1:x)
names(sn) <- c(1:x)

model1 <- tnam(
diffs ~
covariate(wt, coefname = "wt") +
weightlag(diffs, tnet) +
centrality(tnet, type = "indegree") +
centrality(tnet, type = "outdegree") +
clustering(tnet, directed = TRUE) +
degreedummy(tnet, deg = 0, type = c("indegree")) +
degreedummy(tnet, deg = 0, type = c("outdegree"))
)
summary(model1)

from tnam.

The problem is that your wt covariate is identical with the outdegree, thereby causing an identification problem. At least for the first time point. After that, the covariate is NA, which probably doesn't help because all those other time points never enter the model because they are dropped due to the NA values. You can inspect your data structure by either running the model terms as separate functions or using tnamdata instead of tnam in your function call.

from tnam.