simulate_selfreport_network: Simulate a self-reported network

In ctross/STRAND: An R package for simulating and analyzing social network data in R and Stan

Simulate a self-reported network

Description

This function allows the user to simulate a self-reported network, along with a 'true' network, and a network of resource flows. The function first simulates the true network data using the simulate_sbm_plus_srm_network() function, and then simulate self-reports and observable resource flows over the true network. This allows the user to investigate the effects of response biases, such as false positive rate, on network properties.

Usage

simulate_selfreport_network(
  N_id = 99,
  B = NULL,
  V = 3,
  groups = NULL,
  sr_mu = c(0, 0),
  sr_sigma = c(1, 1),
  sr_rho = 0.6,
  dr_mu = c(0, 0),
  dr_sigma = 1,
  dr_rho = 0.7,
  individual_predictors = NULL,
  dyadic_predictors = NULL,
  individual_effects = NULL,
  dyadic_effects = NULL,
  fpr_effects = NULL,
  rtt_effects = NULL,
  theta_effects = NULL,
  false_positive_rate = c(0.01, 0.02, 0),
  recall_of_true_ties = c(0.8, 0.6, 0.99),
  theta_mean = 0.125,
  fpr_sigma = c(0.3, 0.2, 0),
  rtt_sigma = c(0.5, 0.2, 0),
  theta_sigma = 0.2,
  N_responses = 2,
  N_periods = 12,
  flow_rate = rbeta(12, 3, 30),
  decay_curve = rev(1.5 * exp(-seq(1, 4, length.out = 12)))
)

Arguments

N_id	Number of individuals.
B	List of matrices that hold intercept and offset terms. Log-odds. The first matrix should be 1 x 1 with the value being the intercept term.
V	Number of blocking variables in B.
groups	Dataframe of the block IDs of each individual for each variable in B.
sr_mu	Mean vector for sender and receivier random effects. In most cases, this should be c(0,0).
sr_sigma	Standard deviation vector for sender and receivier random effects. The first element controls node-level variation in out-degree, the second in in-degree.
sr_rho	Correlation of sender-receiver effects (i.e., generalized reciprocity).
dr_mu	Mean vector for dyadic random effects. In most cases, this should be c(0,0).
dr_sigma	Standard deviation for dyadic random effects.
dr_rho	Correlation of dyad effects: (i.e., dyadic reciprocity).
individual_predictors	An N_id by N_individual_parameters matrix of covariates.
dyadic_predictors	An N_id by N_id by N_dyadic_parameters matrix of covariates.
individual_effects	A 2 by N_individual_parameters matrix of slopes. The first row gives effects of focal characteristics (on out-degree). The second row gives effects of target characteristics (on in-degree).
dyadic_effects	An N_dyadic_parameters vector of slopes.
fpr_effects	A 3 by N_predictors matrix of slopes. The first row controls the effects of covariates on layer 1 responses, and the second row layer 2 responses. The third row controls the effects of covariates on false positive in the observation network.
rtt_effects	A 3 by N_predictors matrix of slopes. The first row controls the effects of covariates on layer 1 responses, and the second row layer 2 responses. The third row controls the effects of covariates on false positive in the observation network.
theta_effects	An N_predictors vector of slopes controling the effects of covariates on name duplication from layer 1 to layer 2.
false_positive_rate	The baseline false positive rate. This should be supplied as a 3 vector, with each element controlling the false positive rate for a single layer. Support is on the unit interval. If covariates are centered, this can be loosly thought of as an average false positive rate.
recall_of_true_ties	The baseline recall rate of true ties. This should be supplied as a 3 vector, with each element controlling the true tie recall rate for a single layer. Support is on the unit interval. If covariates are centered, this can be loosly thought of as an average true tie recall rate.
theta_mean	The baseline probability of name duplication from layer 1 to layer 2. Scalar. Support is on the unit interval. If covariates are centered, this can be loosly thought of as an average true tie recall rate.
fpr_sigma	Standard deviation 3-vector for false_positive_rate random effects. There should be one value for each layer.
rtt_sigma	Standard deviation 3-vector for recall_of_true_ties random effects. There should be one value for each layer.
theta_sigma	Standard deviation scalar for theta random effects.
N_responses	Number of self-report layers. =1 for single-sampled, =2 for double-sampled.
N_periods	Number of time-periods in which observed transfers are sampled.
flow_rate	A vector of length N_periods, each element controls the probability that a true tie will result in a transfer in a given time period.
decay_curve	A vector of length N_periods, each element controls the increment log-odds of recalling a true tie at time T, as a function of a transfer occuring in period t.

Value

A list of data formatted for use in Stan models.

Examples

## Not run: 
library(igraph)
V = 1            # One blocking variable
G = 3            # Three categories in this variable
N_id = 100       # Number of people

clique = sample(1:3, N_id, replace=TRUE)
B = matrix(-8, nrow=G, ncol=G)
diag(B) = -4.5 # Block matrix

B[1,3] = -5.9
B[3,2] = -6.9

A = simulate_selfreport_network(N_id = N_id, B=list(B=B), V=V, 
                         groups=data.frame(clique=factor(clique)),
                         individual_predictor=matrix(rnorm(N_id, 0, 1), nrow=N_id, ncol=1), 
                         individual_effects=matrix(c(1.7, 0.3),ncol=1, nrow=2),
                         sr_sigma = c(1.4, 0.8), sr_rho = 0.5,
                         dr_sigma = 1.2, dr_rho = 0.8,
                         false_positive_rate = c(0.00, 0.00, 0.00), 
                         recall_of_true_ties = c(0.7, 0.3, 0.99),
                         theta_mean = 0.0, 
                         fpr_sigma = c(0.0, 0.0, 0.0), 
                         rtt_sigma = c(0.5, 0.2, 0.0),
                         theta_sigma = 0.0,
                         N_responses = 2,
                         N_periods = 1, 
                         flow_rate = rbeta(1, 3, 30),
                         decay_curve = rep(0,1)
                         )

par(mfrow=c(1,3))

# True Network
Net = graph_from_adjacency_matrix(A$true_network, mode = c("directed"))
V(Net)$color = c("turquoise4","gray13", "goldenrod3")[A$group_ids$clique]

plot(Net, edge.arrow.size =0.1, edge.curved = 0.3, vertex.label=NA, vertex.size = 5)


# Reported - out transfers
Net = graph_from_adjacency_matrix(A$reporting_network[,,1], mode = c("directed"))
V(Net)$color = c("turquoise4","gray13", "goldenrod3")[A$group_ids$clique]

plot(Net, edge.arrow.size =0.1, edge.curved = 0.3, vertex.label=NA, vertex.size = 5)


# Reported - in transfers
Net = graph_from_adjacency_matrix(A$reporting_network[,,2], mode = c("directed"))
V(Net)$color = c("turquoise4","gray13", "goldenrod3")[A$group_ids$clique]

plot(Net, edge.arrow.size =0.1, edge.curved = 0.3, vertex.label=NA, vertex.size = 5)

## End(Not run)

ctross/STRAND documentation built on Nov. 14, 2024, 11:50 p.m.