README.md

In mihaiconstantin/netPower: Simulations for the master's thesis

This repository has been moved and it is no longer maintained. Please read below.

This repository is part of the research master thesis titled Sample Size Recommendations for Estimating Cross-Sectional Network Models by Mihai A. Constantin, under the supervision of Angélique O. J. Cramer. In this project we performed simulations for several network models to provide sample size recommendations. The primary purpose of this repository is to allow researchers to replicate the results presented in the thesis (i.e., using the repository version at commit 9ca8392). The thesis and the simulation scripts are freely available on OSF at osf.io/zkaxu.

If you are interested in sample size analysis for psychological networks, you are cordially invited to take a look at following paper:

• Constantin, M. A., Schuurman, N. K., & Vermunt, J. (2021). A General Monte Carlo Method for Sample Size Analysis in the Context of Network Models. PsyArXiv. https://doi.org/10.31234/osf.io/j5v7u

In this paper, which we submitted for publication, we introduce a general method to perform sample size calculations for psychological networks. To facilitate researchers to conduct such sample size calculations we provide an R package called powerly available both on GitHub at mihaiconstantin/powerly and CRAN. The package powerly is stable and under active development to support new models. To request a new model, performance measure, or statistic, please open an issue at github.com/mihaiconstantin/powerly/issues.

netPower

Sample Size Recommendations for Estimating Cross-Sectional Network Models.

(see github.com/mihaiconstantin/powerly)

Installation

• the package can be installed as devtools::install_github('mihaiconstantin/netPower')

Description

The data used in the study can be loaded via data("netPowerData"). The code block below illustrates the main four functions found in this package.

1. build_design(participants, nodes, architectures, connectedness, models)

• builds the simulation design

# Specifying the factors.
participants    = seq(50, 1000, 50)
nodes           = c(10, 20, 30)
architectures   = c(random = 1, small_world = 2, scale_free = 3)
connectedness   = c(low = 1, medium = 2, large = 3)
models          = c(ising = 1, ggm = 2)

# Building the design.
design = build_design(participants, 
                      nodes, 
                      architectures, 
                      connectedness,
                      models)

head(design)
#   participants nodes architectures connectedness models
# 1           50    10             1             1      1
# 2           50    10             1             1      2
# 3           50    10             1             2      1
# 4           50    10             1             2      2
# 5           50    10             1             3      1
# 6           50    10             1             3      2

2. run_cell(participants, nodes, density, architecture)

• applies the simulation procedure on a single design cell (i.e., referred to as row in the design matrix above)

result_cell = run_cell(participants  = 300,
                       nodes         = 10, 
                       architecture  = 2,
                       connectedness = 3,
                       model         = 2)

result_cell
#  Simulation cell results:  
#       - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 
#    -> config: 300 participants | 10 nodes | small world architecture | 3 connectedness | ggm model. 
#       - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 
#    -> sensitivity: 0.85 
#    -> specificity: 0.88 
#    -> type one error: 0.12 
#    -> type two error: 0.15 
#    -> edge correlation: 0.9335286 
#    -> equal # nodes: yes 
# Try plot(result) for a visual inspection.

plot(result_cell)

• the output contains a list containing two sub-lists: $raw, $computed:
  • $raw contains the true and estimated model parameters and the data
  • $computed contains the outcome measures of interests

3. run_cells(cells)

• applies the simulation procedure on a selected number of cells (i.e., it is a wrapper around run_cell)

result_cells = run_cells(design[c(1, 97, 165), ])
# -> Running simulation for 3 cells:
#   -> config: 50 par | 10 nod | 1 arc | 1 con | 1 mod. Cell done. ✓
#   -> config: 100 par | 30 nod | 2 arc | 1 con | 1 mod. Cell done. ✓
#   -> config: 200 par | 10 nod | 1 arc | 2 con | 1 mod. Cell done. ✓
# -> Completed all 3 cells.

4. replicate_cells(cells, replications)

• applies the simulation procedure on a selected number of cell and replicates the procedure n number of times (i.e., it is a wrapper around run_cells)

replicated cells = replicate_cells(design[c(1, 97, 165), ], 2)
# Design replications requested: 2.

# ------------------------------
# Replication: 1.
# ------------------------------
# -> Running simulation for 3 cells:
#   -> config: 50 par | 10 nod | 1 arc | 1 con | 1 mod. Cell done. ✓
#   -> config: 100 par | 30 nod | 2 arc | 1 con | 1 mod. Cell done. ✓
#   -> config: 200 par | 10 nod | 1 arc | 2 con | 1 mod. Cell done. ✓
# -> Completed all 3 cells.

# ------------------------------
# Replication: 2.
# ------------------------------
# -> Running simulation for 3 cells:
#   -> config: 50 par | 10 nod | 1 arc | 1 con | 1 mod. Cell done. ✓
#   -> config: 100 par | 30 nod | 2 arc | 1 con | 1 mod. Cell done. ✓
#   -> config: 200 par | 10 nod | 1 arc | 2 con | 1 mod. Cell done. ✓
# -> Completed all 3 cells.

# Completed all 2 replications.

Note: For more details please check the documentation of each function (i.e., work in progress).

mihaiconstantin/netPower documentation built on Oct. 4, 2021, 10:16 p.m.