Contains model-based treatment of missing data for regression
models with missing values in covariates or the dependent
variable using maximum likelihood or Bayesian estimation
(Ibrahim et al., 2005;
If you use mdmb and have suggestions for improvement or have found bugs, please email me at [email protected].
The official version of mdmb is hosted on CRAN and may be found here.
The CRAN version can be installed from within R using:
utils::install.packages("mdmb")The version hosted here is the development version of mdmb.
The GitHub version can be installed using devtools as:
devtools::install_github("alexanderrobitzsch/mdmb")+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
------------------------- mdmb NEWS ---------------------------
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
|\ /||\ |\ /||
| / || || / ||--<
| ||/ | ||/
mdmb: Model Based Treatment of Missing Data
Alexander Robitzsch & Oliver Luedtke
Questions or suggestions about mdmb should be sent to [email protected]
In case of reporting a bug, please always provide a reproducible script.
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
CHANGELOG mdmb
NOTE * included consistency checks of model names for covariate models in frm_em() and frm_fb() NOTE * fixed an error in mdmb_regression() for Yeo-Johnson transformed variables (incompatible dimensions)
DATA * included/modified datasets: --- EXAMP * included/modified examples: ---
NOTE * included argument 'control_optim_fct' in oprobit_regression(). Slight speed improvements in case of many ordered categories. FIXED * fixed a recently introduced bug for linear regression models in frm_fb()
DATA * included/modified datasets: --- EXAMP * included/modified examples: ---
ADDED * extended Yeo-Johnson transformation to include bounded variables on [0,1] by employing a probit transformation. The extension is implemented in fit_yjt_scaled(), yjt_regression(), frm_em() and frm_fb().
DATA * included/modified datasets: data.mb05 EXAMP * included/modified examples: yjt_dist (6), frm_em (11)
ADDED * included option for Gibbs sampling in frm_fb() which can be enabled by the argument 'use_gibbs' ADDED * included multilevel regression for normally distributed and ordinal data in frm_fb() (model 'mlreg')
DATA * included/modified datasets: data.mb04 EXAMP * included/modified examples: frm (10)
NOTE * included argument 'as_mids' for conversion into mids objects in frm2datlist() NOTE * translated some parts of calculations in frm_em() into Rcpp
DATA * included/modified datasets: --- EXAMP * included/modified examples: ---
FIXED * fixed bug in oprobit_regression() when no regression coefficients were estimated in the model FIXED * fixed incorrect matching of MCMC diagnostics informations in summary output FIXED * included correct labelling in frm_em() and frm_fb() if classes 'yjtreg' and 'bctreg' are used NOTE * included dataset data.mb03 from Enders et al. (2014, PsychMeth) FIXED * fixed problems in numerical overflow when calculating Metropolis-Hasting ratios in frm_fb()
DATA * included/modified datasets: data.mb03 EXAMP * included/modified examples: data.mb (1)
ADDED * added example for modelling nonignorable data with frm_em() (see Example 8 in 'frm.Rd') ADDED * added ordinal probit distribution fit_oprobit() and ordinal probit regression oprobit_regression(). ADDED * extended frm_em() and frm_fb() functions to ordinal data NOTE * fixed incorrect variable names in parts of the MCMC summary output in frm_fb()
DATA * included/modified datasets: --- EXAMP * included/modified examples: frm (8), oprobit_dist (1), mdmb_regression (4), frm (9)
NOTE * speeded data processing in frm_em()
NOTE * included Example 1.3 in frm() using the jomo package for
imputation under a substantive model containing interaction
effects
NOTE * included example for estimation of model including latent
interaction effects with frm_em() function
NOTE * included argument 'log' in dt_scaled(), dbct_scaled() and
dyt_scaled()
NOTE * included more efficient computation of gradient in
logistic_regression(), bct_regression() and yjt_regression().
Different computation methods can be chosen by the argument
'use_grad'. These gradients are now also used in frm_em()
which results in some speed improvement.
NOTE * changed initial values in logistic_regression(),
bct_regression() and yjt_regression() to least-squares
solutions
DATA * included/modified datasets: --- EXAMP * included/modified examples: frm (1.3, 7)
NOTE * changed default in 'nodes_control' in 'frm_em' to use a wider integration grid for missing values NOTE * added information about standard error calculation in 'frm_em'; reference Jamshidian and Jennrich (2000, JRSSB) FIXED * fixed an error in the 'frm_em' function which was not yet applicable for 'yjtreg' regression models
DATA * included/modified datasets: --- EXAMP * included/modified examples: ---
FIXED * fixed an error in initialization of sigma parameter in mdmb::frm_fb function with 'linreg' imputation
DATA * included/modified datasets: --- EXAMP * included/modified examples: ---
INIT * initial version of the package ADDED * added functions for scaled t distribution with Yeo-Johnson transformation ('d_yjt_scaled', 'fit_yjt_scaled') and Box-Cox transformation ('d_bct_scaled', 'fit_bct_scaled') ADDED * added additional regression functions 'logistic_regression', 'yjt_regression' and 'bct_regression' ADDED * added function 'frm_em' for maximum likelihood estimation of regression models with missing covariates ADDED * added function 'frm_fb' for fully Bayesian estimation of regression models with missing covariates. Imputations of missing values are provided. NOTE * included utility functions 'eval_prior_list', 'eval_prior_list_sumlog', 'offset_values_extract' and 'remove_NA_data_frame'
DATA * included/modified datasets: data.mb01, data.mb02 EXAMP * included/modified examples: mbmb_regression (1,2,3,4,5), frm (1,2,3,4,5), yjt_dist (1,2,3,4,5)