raffalib

A personal R package collecting helper functions the author reaches for again and again in empirical research: building publication-ready tables and exporting them to Word, quick descriptive statistics and data-wrangling shortcuts, list/vector/data-frame utilities, formula manipulation, and a set of helpers around glmmTMB (alternative optimizers and control-function 2SLS).

It is a grab-bag of conveniences rather than a single-purpose package; pick the pieces you need. Everything is documented — see ?raffalib and the individual help pages.

Installation

# install.packages("remotes")
remotes::install_github("raffaelemancuso/raffalib-r")

The package leans on a fairly wide set of CRAN packages (it Imports modelsummary, gtsummary, flextable, officer, glmmTMB, optimx, calibrar, and others); installing from GitHub will pull them in.

What's inside

Tables and Word export

Function	Purpose
correlation_table()	Correlation matrix as a flextable, with a numbered variable legend in the footer.
flextable2docx(), ggplot2docx(), plot2docx()	Write a flextable, a ggplot, or a base R plot to a .docx file, with caption and page-geometry control.
modelsummary_build_labelled_coef_map()	Turn variable labels (including factor levels) into a coef_map for modelsummary().
modelsummary_common_coefs_at_bottom()	Re-order coefficients so terms shared across models print last.
modelsummary_getgofmap(), modelsummary_missing_variables_in_coef_map()	Goodness-of-fit map helper and coef-map diagnostics.
gtsummary_add_mean_diff(), gtsummary_add_significance_stars(), gtsummary_format_statistic_column()	Extensions for gtsummary tables: between-group differences, custom significance stars, statistic-column formatting.

Descriptive statistics

myfreq() (frequency table that always shows NAs), descquant() (evenly spaced quantiles), descstrange() (counts of NA/NaN/Inf), na_per_group(), dupsa() (assert that grouping variables uniquely identify rows), and get_dropped_obs() (the rows a model dropped through listwise deletion).

Data wrangling

nan2na() (replace NaN with NA in numeric columns), sort_columns_by_label() / sort_columns_by_name(), catcols() (print matching column names), and startlog() / endlog() to bracket a pipeline and report how many rows, columns or cells it changed.

Lists and vectors

list_rename_names() / list_rename_values(), sort_named_list_by_names() / sort_named_list_by_values(), catvec(), as_named_list(), vec_relocate().

Formulas

reformulas_addints() (add treatment-by-control interactions) and reformulas_randint() (strip random slopes, keep random intercepts).

glmmTMB helpers

• glmmTMB_2sls() — control-function 2SLS with a (possibly non-Gaussian, possibly mixed) glmmTMB second stage. For a Gaussian second stage it reproduces textbook 2SLS; for count/binary outcomes it is the consistent alternative to the "forbidden regression". Returns an object with print/tidy/glance/nobs methods, so it flows straight into modelsummary(), and offers cluster-bootstrap standard errors that correct for the generated regressor.
• Alternative optimizers — three families of glmmTMBControl() constructors that swap in optimizers from optimx and calibrar when the default nlminb() struggles to converge: glmmTMB_control_optimx_*() (e.g. _bfgs, _nlminb, _bobyqa), glmmTMB_control_calibrar_*() (local methods), glmmTMB_control_optimh_*() (global/heuristic methods such as CMA-ES, differential evolution, particle swarm).
• glmmTMB_get_optimum(), glmmTMB_get_hessian_1() / glmmTMB_get_hessian_2() — extract the current estimates (to warm-start a refit) and the Hessian at the optimum.

Miscellaneous

find_method() (which S3 method a generic would dispatch), gen_batches() (split an index range into chunks), and save_backup() / read_backup() (time-stamped .rds snapshots).

Examples

Save a regression table to Word:

library(raffalib)
library(modelsummary)

mod <- lm(mpg ~ wt + hp, data = mtcars)
tbl <- modelsummary(mod, output = "flextable")
flextable2docx(tbl, "regression.docx", word_prop = list(caption_text = "Table 1"))

Fit an IV model whose second stage is a Poisson glmmTMB:

m <- glmmTMB_2sls(
  first_stage  = x ~ z + w,   # endogenous x, instrument z, control w
  second_stage = y ~ x + w,
  data = d, family = poisson(),
  instruments = "z", n_boot = 200
)
m                 # weak-instrument and Wu-Hausman diagnostics
generics::tidy(m) # ready for modelsummary()

Refit a stubborn model with a different optimizer:

glmmTMB::glmmTMB(
  count ~ mined + (1 | site),
  family = poisson, data = glmmTMB::Salamanders,
  control = glmmTMB_control_optimx_bfgs()
)

License

GPL-3. See LICENSE.txt.