NEWS

medflex 0.6-10 (2023-06-22)

Issue related to changes to residualPlot.neModel and residualPlots.neModel (in version 0.6-9) fixed (special thanks to Yves Rosseel for diagnosing and fixing)

Errors and warnings in residualPlot.expData, residualPlots.expData, residualPlot.neModel and residualPlots.neModel (due to changes in the car package) fixed (thanks to Sandy Weisberg for diagnosing and helping out)

Issue with neWeight function using VGAM::vglm with family = multinomial solved (cf github issue #15)
neWeight function and xFit argument in neModel function now compatible with family = uninormal specification (VGAM::vglm function) ("gaussianff" was deprecated)

neEffdecomp no longer fails when covariate name is substring of outcome name (cf github issue #13)

First byte-compiled version
Correction of example in neImpute help files with binary exposure (should be explicitly coded as factor!)
Issue with bootstrap and data.frame named 'data' solved (github issue #8)
Issue with family specification for propensity score model solved (github issue #11)
Issue with special terms of formula of vgam function solved (github issue #12)

Error when running neEffdecomp on a neModel-object that represents a population-average natural effect model: fixed (cf github issue #7)

Added flexibility with formula argument (in order to mimick glm function's behavior) (cf github issue #2)
Compatibility with tbl_df function from dplyr package (cf github pull request #3)
Bug fixed when using vglm function from VGAM package in (i) neImpute: returned imputed outcomes on scale of link function instead of mean outcome scale (use of predictvglm instead of predict.vglm in order for type = "response" to be properly specified), (ii) neWeight and (iii) neModelEst (when specifying xFit): either returned incorrect weights or error (same reason as for neImpute) (cf github issue #4)

Compatibility with mice and mitools packages for multiple imputation (see vignette for documentation and example)
Sandwich estimator now also accommodates settings with missingness only in the outcome (requires MAR assumption for imputation-based approach on all cases and MCAR assumption for weighting-based approach on complete cases): no more error returned
Goodness-of-fit tools added: convenience functions residualPlot and residualPlots from the car package can now also be used on expData-class and neModel-class objects to assess model adequacy of the working model(s) and natural effect models, respectively.
Error when name of the mediator variable is a substring of the name of the exposure variable ("The original mediator variables should not be included in the natural effect model!"): fixed
Warning about model uncongeniality when name of the exposure variable is a substring of the name of the mediator variable (cf mail correspondence with CY 08/07/2015): bug fixed

Error/bug when using factor() terms in neEffdecomp: fixed
neWeight now returns a warning when trying to specify more than one mediator (e.g. 'nMed = 2'), as joint mediated effects are only implemented for the imputation-based approach.
Coding error for outcome variable 'UPB' in UPBdata dataset: fixed (up to version 0.5-0 'UPB' was a binary variable indicating whether the individual reported having displayed at least 4 events of unwanted pursuit behavior; while it stated to indicate whether the individual reported having displayed any unwanted pursuit behavior(s) towards the ex-partner)

Robust standard errors based on the sandwich estimator as alternative for bootstrap SEs (option 'se = "robust"' in neModel function)
Advanced options for effect decomposition (neEffdecomp function now allows to specify reference exposure levels and covariate levels via 'xRef' and 'covLev' arguments, respectively)
Additional option to weight for multicategorical variables (either via inverse-probability-of-treatment weighting for multivariate exposures or via ratio-of-mediator-probability weighting for multivariate mediators) using 'vglm' from the 'VGAM' package as model fitting function
Vignettes with more details added
Ratio-of-mediator-probability weights can be extracted from the expanded dataset object via 'weights' function
Minor changes to UPBdata (different versions of variables + variable 'initiator' added)
Bugs in neEffdecomp fixed: error returned if imputation model did not contain covariates, 'ci.type' argument can now be specified
Bug in Anova() applied to neModel object fixed