Example High-Dimensional Mediation Data Sets

Three small simulated data sets, one per outcome type, for trying the POEMED tests and reproducing the help-page examples. Each was generated by simulate_mediation_data() under a homogeneous mediation pattern (a few same-sign active mediators among many null ones) with three confounders, at seed 113, so the generating truth is known and travels with the data.

Usage

example_continuous
example_binary
example_count

Format

Each is a list, the return value of simulate_mediation_data(), with components:

X

Exposure matrix, \(n \times 1\).

M

Candidate mediator matrix, \(n \times 100\).

Y

Outcome vector of length \(n\): continuous for example_continuous, 0/1 for example_binary, nonnegative counts for example_count.

Z

Confounder matrix, \(n \times 3\).

alpha_m, Gamma_x, alpha_x

The mediator-on-outcome, exposure-on-mediator, and direct-effect coefficients used to generate the data.

alpha_z, Gamma_z

The confounder coefficients the simulator drew from U(-0.5, 0.5) and used: alpha_z, the three confounder-on-outcome coefficients, and Gamma_z, the \(3 \times 100\) matrix of confounder-on-mediator coefficients, in which each confounder loads equally on every mediator (each row holds one value).

beta

The total indirect effect \(\Gamma_x \alpha_m\): 0.42 for example_continuous, 0.8 for example_binary, and 0.24 for example_count.

active_mediators

Indices of the truly active mediators: mediators 1 to 5 in example_continuous, and mediators 1, 2, and 5 in example_binary and example_count.

n, p, q, outcome, pattern

The generating settings.

The sample sizes are \(n = 200\) (example_continuous) and \(n = 300\) (example_binary, example_count).

In example_count the simulator's overflow guard fired (the largest log-mean was 5.88; see the Value section of simulate_mediation_data()), so the counts follow a log-mean whose coefficients are the recorded alpha_m, alpha_x, and alpha_z multiplied by 0.71 (with an intercept of 0.83), and a total indirect effect of 0.17 where beta records 0.24.

Source

Simulated with simulate_mediation_data(); see data-raw/example_data.R in the package sources.

See also

pe_mediation(), simulate_mediation_data().

Examples

data(example_continuous)
str(example_continuous$active_mediators)   # truly active: mediators 1 to 5
#>  int [1:5] 1 2 3 4 5
fit <- pe_mediation(example_continuous$X, example_continuous$Y,
                    example_continuous$M, Z = example_continuous$Z,
                    outcome = "continuous")
# Both tests reject at the 0.05 level, and the screen identifies one of
# the five truly active mediators (mediator 4).
fit
#>  term                  value   
#>  stat_hdmm             6.146   
#>  pval_hdmm             0.0132  
#>  stat_pe               482.6   
#>  j_pe                  476.5   
#>  pval_pe               < 0.0001
#>  total_indirect_effect 0.3758  
#>  n_active_mediators    1       
#>  df                    1       
#>  n_candidate_mediators 100     
#>  n_observations        200     
#> 
#> Outcome model: continuous (linear)
#> Active mediators identified (1): 4
#> Tuning parameter (HBIC): lambda = 0.151 from 100 values in [0.05, 10]