powerly: Perform sample size analysis

In powerly: Sample Size Analysis for Psychological Networks and More

Perform sample size analysis

Description

Run an iterative three-step Monte Carlo method and return the sample sizes required to obtain a certain value for a performance measure of interest (e.g., sensitivity) given a hypothesized network structure.

Usage

powerly(
  range_lower,
  range_upper,
  samples = 30,
  replications = 30,
  model = "ggm",
  ...,
  model_matrix = NULL,
  measure = "sen",
  statistic = "power",
  measure_value = 0.6,
  statistic_value = 0.8,
  monotone = TRUE,
  increasing = TRUE,
  spline_df = NULL,
  solver_type = "quadprog",
  boots = 10000,
  lower_ci = 0.025,
  upper_ci = 0.975,
  tolerance = 50,
  iterations = 10,
  cores = NULL,
  backend_type = NULL,
  save_memory = FALSE,
  verbose = TRUE
)

Arguments

range_lower	A single positive integer representing the lower bound of the candidate sample size range.
range_upper	A single positive integer representing the upper bound of the candidate sample size range.
samples	A single positive integer representing the number of sample sizes to select from the candidate sample size range.
replications	A single positive integer representing the number of Monte Carlo replications to perform for each sample size selected from the candidate range.
model	A character string representing the type of true model to find a sample size for. Possible values are "ggm" (the default).
...	Required arguments used for the generation of the true model. See the True Models section for the arguments required for each true model.
model_matrix	A square matrix representing the true model. See the True Models section for what this matrix should look like depending on the true model selected.
measure	A character string representing the type of performance measure of interest. Possible values are "sen" (i.e., sensitivity; the default), "spe" (i.e., specificity), "mcc" (i.e., Matthews correlation), and "rho" (i.e., Pearson correlation). See the True Models section for the performance measures available for each type of true model supported.
statistic	A character string representing the type of statistic to be computed on the values obtained for the performance measures. Possible values are "power" (the default).
measure_value	A single numerical value representing the desired value for the performance measure of interest. The default is 0.6 (i.e., for the measure = "sen"). See the Performance Measures section for the range of values allowed for each performance measure.
statistic_value	A single numerical value representing the desired value for the statistic of interest. The default is 0.8 (i.e., for the statistic = "power"). See the "Statistics" section for the range of values allowed for each statistic.
monotone	A logical value indicating whether a monotonicity assumption should be placed on the values of the performance measure. The default is TRUE meaning that the performance measure changes as a function of sample size (i.e., either by increasing or decreasing as the sample size goes up). The alternative FALSE indicates that the performance measure it is not assumed to change as a function a sample size.
increasing	A logical value indicating whether the performance measure is assumed to follow a non-increasing or non-decreasing trend. TRUE (the default) indicates a non-decreasing trend (i.e., the performance measure increases as the sample size goes up). FALSE indicates a non-increasing trend (i.e., the performance measure decreases as the sample size goes up).
spline_df	A vector of positive integers representing the degrees of freedom considered for constructing the spline basis, or NULL. The best degree of freedom is selected based on Leave One Out Cross-Validation. If NULL (the default) is provided, a vector of degrees of freedom is automatically created with all integers between 3 and 20.
solver_type	A character string representing the type of the quadratic solver used for estimating the spline coefficients. Possible values are "quadprog" (the default) and "osqp". Currently, the "osqp" solver does not play nicely with R's parallel::parallel package and cannot be used when powerly is ran in parallel.
boots	A positive integer representing the number of bootstrap runs to perform on the matrix of performance measures in order to obtained bootstrapped values for the statistic of interest. The default is 10000.
lower_ci	A single numerical value indicating the lower bound for the confidence interval to be computed on the bootstrapped statistics. The default is 0.025 (i.e., 2.5%).
upper_ci	A single numerical value indicating the upper bound for the confidence to be computed on the bootstrapped statistics. The default is 0.975 (i.e., 97.5%).
tolerance	A single positive integer representing the width at the candidate sample size range at which the algorithm is considered to have converge. The default is 50, meaning that the algorithm will stop running when the difference between the upper and the lower bound of the candidate range shrinks to 50 sample sizes.
iterations	A single positive integer representing the number of iterations the algorithm is allowed to run. The default is 10.
cores	A single positive positive integer representing the number of cores to use for running the algorithm in parallel, or NULL. If NULL (the default) the algorithm will run sequentially.
backend_type	A character string indicating the type of cluster to create for running the algorithm in parallel, or NULL. Possible values are "psock" and "fork". If NULL the backend is determined based on the computer architecture (i.e., fork for Unix and MacOS and psock for Windows).
save_memory	A logical value indicating whether to save memory by only storing the results for the last iteration of the method. The default TRUE indicates that only the last iteration should be saved.
verbose	A logical value indicating whether information about the status of the algorithm should be printed while running. The default is TRUE.

Details

This function represents the implementation of the method introduced by Constantin et al. (2021; see doi: 10.31234/osf.io/j5v7u) for performing a priori sample size analysis in the context of network models. The method takes the form of a three-step recursive algorithm designed to find an optimal sample size value given a model specification and an outcome measure of interest (e.g., sensitivity). It starts with a Monte Carlo simulation step for computing the outcome of interest at various sample sizes. It continues with a monotone non-decreasing curve-fitting step for interpolating the outcome. The final step employs a stratified bootstrapping scheme to account for the uncertainty around the recommendation provided. The method runs the three steps recursively until the candidate sample size range used for the search shrinks below a specified value.

Value

An R6::R6Class() instance of Method class that contains the results for each step of the method for the last and previous iterations.

Main fields:

• $duration: The time in seconds elapsed during the method run.

• $iteration: The number of iterations performed.

• $converged: Whether the method converged.

• $previous: The results during the previous iteration.

• $range: The candidate sample size range.

• $step_1: The results for Step 1.

• $step_2: The results for Step 2.

• $step_3: The results for Step 3.

• $recommendation: The sample size recommendation(s).

The plot method can be called on the return value to visualize the results. See plot.Method() for more information on how to plot the method results.

• for Step 1: plot(results, step = 1)

• for Step 2: plot(results, step = 2)

• for Step 3: plot(results, step = 3)

True Models

Gaussian Graphical Model (GGM)

• type: cross-sectional

• symbol: ggm

• ... arguments for generating true models:

  • nodes: A single positive integer representing the number of nodes in the network (e.g., 10).

  • density: A single numerical value indicating the density of the network (e.g., 0.4).

  • positive: A single numerical value representing the proportion of positive edges in the network (e.g., 0.9 for 90% positive edges).

  • range: A length two numerical value indicating the uniform interval from where to sample values for the partial correlations coefficients (e.g., c(0.5, 1)).

  • constant: A single numerical value representing the constant described by Yin and Li (2011).

  • for more information on the arguments see:

    • the function bootnet::genGGM()

    • Yin, J., and Li, H. (2011). A sparse conditional gaussian graphical model for analysis of genetical genomics data. The annals of applied statistics, 5(4), 2630.

• supported performance measures: sen, spe, mcc, rho

Performance Measures

Performance Measure	Symbol	Lower	Upper
Sensitivity	sen	0.00	1.00
Specificity	spe	0.00	1.00
Matthews correlation	mcc	-1.00	1.00
Pearson correlation	rho	-1.00	1.00

Statistics

Statistics	Symbol	Lower	Upper
Power	power	0.00	1.00

Requests

• If you would like to support a new model, performance measure, or statistic, please open a pull request on GitHub at github.com/mihaiconstantin/powerly/pulls.

• To request a new model, performance measure, or statistic, please submit your request at github.com/mihaiconstantin/powerly/issues. If possible, please also include references discussing the topics you are requesting.

• Alternatively, you can get in touch at mihai at mihaiconstantin dot com.

References

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. doi: 10.31234/osf.io/j5v7u

See Also

plot.Method(), summary.Method(), validate(), generate_model()

Examples

# Suppose we want to find the sample size for observing a sensitivity of `0.6`
# with a probability of `0.8`, for a GGM true model consisting of `10` nodes
# with a density of `0.4`.

# We can run the method for an arbitrarily generated true model that matches
# those characteristics (i.e., number of nodes and density).
results <- powerly(
    range_lower = 300,
    range_upper = 1000,
    samples = 30,
    replications = 30,
    measure = "sen",
    statistic = "power",
    measure_value = .6,
    statistic_value = .8,
    model = "ggm",
    nodes = 10,
    density = .4,
    cores = 2,
    verbose = TRUE
)

# Or we omit the `nodes` and `density` arguments and specify directly the edge
# weights matrix via the `model_matrix` argument.

# To get a matrix of edge weights we can use the `generate_model()` function.
true_model <- generate_model(type = "ggm", nodes = 10, density = .4)

# Then, supply the true model to the algorithm directly.
results <- powerly(
    range_lower = 300,
    range_upper = 1000,
    samples = 30,
    replications = 30,
    measure = "sen",
    statistic = "power",
    measure_value = .6,
    statistic_value = .8,
    model = "ggm",
    model_matrix = true_model,
    cores = 2,
    verbose = TRUE
)

# To visualize the results, we can use the `plot` S3 method and indicating the
# step that should be plotted.
plot(results, step = 1)
plot(results, step = 2)
plot(results, step = 3)

# To see a summary of the results, we can use the `summary` S3 method.
summary(results)

powerly documentation built on Sept. 9, 2022, 5:07 p.m.