backbone (version 2.0.0)

fdsm.trials: Estimate number of Monte Carlo trials needed for FDSM backbone

Description

fdsm.trials estimates the number of Monte Carlo trials needed to extract an FDSM backbone, correcting for familywise error rate, and given tolerance for Type-I and Type-II errors

Usage

fdsm.trials(
  B,
  type1 = 0.05,
  type2 = 0.05,
  alpha = 0.05,
  fwer = TRUE,
  signed = FALSE,
  riskyp = 0
)

Arguments

B

graph: An unweighted bipartite graph object of class matrix, sparse matrix, igraph, edgelist, or network object. Any rows and columns of the associated bipartite matrix that contain only zeros are automatically removed before computations.

type1

numeric: Type-I error used in sample size calculation

type2

numeric: Type-II error used in sample size calculation

alpha

numeric: Desired Type-I error for tests of edge significance

fwer

boolean: If TRUE, alpha is interpreted as the desired familywise error rate. If FALSE, alpha is interpreted as the desired testwise error rate.

signed

boolean: TRUE to estimate the number of trials needed to extract a signed backbone, FALSE to estimate the number of trials needed to extract a binary backbone

riskyp

numeric: Expected riskiest edge p-value, as a proportion of alpha (see details)

Value

integer: estimated minimum number of Monte Carlo trials

Details

This function uses sample size estimation equations 2.22 and 2.24 given by Fleiss et al. (2013).

If fwer = TRUE, it assumes that a conservative Bonferroni correction will be used to maintain the familywise error rate across the independent hypothesis tests required for every edge in the bipartite projection of B.

The required number of trials depends in part on the difference between an edge's estimated p-value and the desired level of statistical significance. If an edge is deemed statistically significant when its p-value is less than 0.05, then there is little risk in making a decision about an edge with an estimated p-value of 0, and fewer trials are required. In contrast, if the edge's estimated p-value is 0.049, there is more risk of making an error and more trials are required. The riskyp parameter specifies how close to alpha the riskiest expected edge p-value, as a proportion of alpha. For example, if alpha = 0.05 and riskyp = 0.75, then the expected riskiest p-value is 0.0375.

References

Fleiss, J. L., Levin, B., & Paik, M. C. (2013). Statistical methods for rates and proportions. John Wiley & Sons.

Neal, Z. P., Domagalski, R., and Sagan, B. (2021). Comparing Alternatives to the Fixed Degree Sequence Model for Extracting the Backbone of Bipartite Projections. Scientific Reports.

Examples

# NOT RUN {
B <- matrix(rbinom(100*1000,1,0.5),100,1000)
fdsm.trials(B, riskyp = .75)
# }