Romeb: Robust Median-Based Bayesian Growth Curve Modeling

Implements robust median-based Bayesian growth-curve models that handle MCAR/MAR/MNAR missing-data mechanisms and complete data. Models are fitted via rjags/JAGS and summarized with coda.

Fits a median-based Bayesian growth curve model under MCAR, MAR, MNAR or complete-data assumptions. If K > 0 the first K columns in data are treated as auxiliary variables.

Usage

Romeb(
  Missing_Type,
  data,
  time,
  seed,
  K = 0,
  chain = 1,
  Niter = 6000,
  burnIn = 3000
)

Arguments

Missing_Type

Character; one of MNAR, MAR, MCAR, no missing.

data

Matrix or data frame containing outcome columns (and optionally auxiliary variables).

time

Numeric vector of measurement times (e.g., c(0,1,2,3)).

seed

Integer seed for reproducibility.

K

Integer; number of auxiliary variables (default 0).

chain

Integer; number of MCMC chains (default 1).

Niter

Integer; iterations per chain (default 6000).

burnIn

Integer; burn-in iterations (default 3000).

Value

An object of class RomebResult containing

quantiles

posterior means, SDs and quantiles

geweke

Geweke z-scores

credible_intervals

95% equal-tail credible intervals

hpd_intervals

95% highest posterior density intervals

samps_full

full coda::mcmc.list (including burn-in)

See also

Useful links:

• https://github.com/DandanTang0/Romeb

• Report bugs at https://github.com/DandanTang0/Romeb/issues

Author

Maintainer: Dandan Tang tangdd20@gmail.com (ORCID)

Authors:

• Xin Tong

Examples

  set.seed(123)
  Y <- matrix(rnorm(300), 100, 3)
  fit <- Romeb("no missing", data = Y, time = c(0,1,2), seed = 123, K = 0,
               Niter = 6000, burnIn = 3000)
#> 
#> Running no missing model  (K = 0)
#> Compiling model graph
#>    Resolving undeclared variables
#>    Allocating nodes
#> Graph information:
#>    Observed stochastic nodes: 300
#>    Unobserved stochastic nodes: 404
#>    Total graph size: 3030
#> 
#> Initializing model
#> 
  print(fit)
#> Romeb GCM summary
#> ==================
#> 
#> Posterior medians (50% quantiles):
#>       par[1]       par[2]       par[3]       par[4]       par[5] 
#> -0.028959076  0.006534935  0.246008195 -0.139386679  0.191886602 
#> 
#> Geweke test:
#>     par[1]     par[2]     par[3]     par[4]     par[5] 
#>  0.6312252 -1.3194087  0.2727626  0.3205222 -0.7499546 
#> 
#> 95% credible intervals:
#>               2.5%       97.5%
#> par[1] -0.20424846  0.14309348
#> par[2] -0.13638706  0.15364752
#> par[3]  0.11134245  0.47052262
#> par[4] -0.31684374 -0.03367711
#> par[5]  0.08970286  0.35341840
#> 
#> 95% hpd intervals:
#>           par[1]     par[2]     par[3]     par[4]     par[5]
#> lower -0.1968473 -0.1370647 0.09913653 -0.2846050 0.07818917
#> upper  0.1489117  0.1522016 0.43895060 -0.0121798 0.33646925
#> 
#> Use x$samps_full to access full MCMC samples, and coda::traceplot(x$samps_full[,'par[i]']) for the trace plot of par[i].