netdiffuseR (version 1.23.0)

toa_mat: Time of adoption matrix

Description

For a single behavior, creates two matrices recording times of adoption of the innovation. One matrix records the time period of adoption for each node with zeros elsewhere. The second records the cumulative time of adoption such that there are ones for the time of adoption and every time period thereafter. For \(Q\) behaviors, creates a list of length \(Q\), where each element contains those two matrices for each behavior.

Usage

toa_mat(obj, labels = NULL, t0 = NULL, t1 = NULL)

Value

For a single behavior, a list of two \(n \times T\):

cumadopt

has 1's for all years in which a node indicates having the innovation.

adopt

has 1's only for the year of adoption and 0 for the rest.

For \(Q\) behaviors, a list of length \(Q\), each element containing

cumadopt ans adopt matrices.

Arguments

obj

Either an integer vector of length \(n\) containing time of adoption of the innovation, a matrix of size \(n \times Q\) (for multiple \(Q\) behaviors), or a diffnet object (both for single or multiple behaviors).

labels

Character vector of length \(n\). Labels (ids) of the vertices.

t0

Integer scalar. Sets the lower bound of the time window (e.g. 1955).

t1

Integer scalar. Sets the upper bound of the time window (e.g. 2000).

Author

George G. Vega Yon, Thomas W. Valente, and Aníbal Olivera M.

Details

In order to be able to work with time ranges other than \(1,\dots, T\) the function receives as input the boundary labels of the time windows through the variables t0 and t. While by default the function assumes that the the boundaries are given by the range of the times vector, the user can set a personalized time range exceeding the one given by the times vector. For instance, times of adoption may range between 2001 and 2005 but the actual data, the network, is observed between 2000 and 2005 (so there is not left censoring in the data), hence, the user could write:

adopmats <- toa_mat(times, t0=2000, t1=2005)

That way the resulting cumadopt and adopt matrices would have 2005 - 2000 + 1 = 6 columns instead of 2005 - 2001 + 1 = 5 columns, with the first column of the two matrices containing only zeros (as the first adoption happend after the year 2000).

For multiple behaviors, the input can be a matrix or a diffnet object. In this case, the output will be a list, with each element replicating the output for a single diffusion: a matrix recording the time period of adoption for each node, and a second matrix with ones from the moment the node adopts the behavior.

Examples

# Random set of times of adoptions
times <- sample(c(NA, 2001:2005), 10, TRUE)

toa_mat(times)

# Now, suppose that we observe the graph from 2000 to 2006
toa_mat(times, t0=2000, t1=2006)

# For multiple behaviors, the input can be a matrix..
times_1 <- c(2001L, 2004L, 2003L, 2008L)
times_2 <- c(2001L, 2005L, 2006L, 2008L)
times <- matrix(c(times_1, times_2), nrow = 4, ncol = 2)

toa <- toa_mat(times)
toa[[1]]$adopt         # time period of adoption for the first behavior

#.. or a diffnet object
graph <- lapply(2001:2008, function(x) rgraph_er(4))
diffnet <- new_diffnet(graph, times)

toa <- toa_mat(diffnet)
toa[[1]]$cumadopt      # cumulative adoption matrix for the first behavior